# Capital One Tech

The low down on our high tech from the engineering experts at Capital One. Learn about the solutions, ideas and stories driving our tech transformation. - Medium

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## A modern dilemma: Consume data from a stream or a lake

DevFeed: [A modern dilemma: Consume data from a stream or a lake](<https://devfeed.tech/articles/a-modern-dilemma-consume-data-from-a-stream-or-a-lake-141161.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/a-modern-dilemma-consume-data-from-a-stream-or-a-lake-296bbb354094?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-11-18T17:35:14Z

Content type: article

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [software-architecture](<https://devfeed.tech/topics/software-architecture.md>), [stream-processing](<https://devfeed.tech/topics/stream-processing.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>), [etl](<https://devfeed.tech/topics/etl.md>), [deltalake](<https://devfeed.tech/topics/deltalake.md>), [Apache Kafka](<https://devfeed.tech/topics/apache-kafka.md>), [Apache Flink](<https://devfeed.tech/topics/apache-flink.md>), [parquet](<https://devfeed.tech/topics/parquet.md>), [Data Quality](<https://devfeed.tech/topics/data-quality.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [apis](<https://devfeed.tech/tags/apis.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [batch](<https://devfeed.tech/tags/batch.md>), [big-data](<https://devfeed.tech/tags/big-data.md>), [business-processes](<https://devfeed.tech/tags/business-processes.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [cloud-computing](<https://devfeed.tech/tags/cloud-computing.md>), [data](<https://devfeed.tech/tags/data.md>), [data-engineering](<https://devfeed.tech/tags/data-engineering.md>), [data-lakes](<https://devfeed.tech/tags/data-lakes.md>), [delta-lake](<https://devfeed.tech/tags/delta-lake.md>), [enrichment](<https://devfeed.tech/tags/enrichment.md>), [etl](<https://devfeed.tech/tags/etl.md>), [event](<https://devfeed.tech/tags/event.md>), [flink](<https://devfeed.tech/tags/flink.md>), [kafka](<https://devfeed.tech/tags/kafka.md>), [latency](<https://devfeed.tech/tags/latency.md>), [legacy](<https://devfeed.tech/tags/legacy.md>), [log](<https://devfeed.tech/tags/log.md>), [mapping](<https://devfeed.tech/tags/mapping.md>), [model-context-protocol](<https://devfeed.tech/tags/model-context-protocol.md>), [software-architecture](<https://devfeed.tech/tags/software-architecture.md>), [software-engineering](<https://devfeed.tech/tags/software-engineering.md>), [stream](<https://devfeed.tech/tags/stream.md>)

### AI overview

Choosing a data consumption pattern early can avoid costly architectural changes later. APIs support synchronous, record-by-record requests and paginated multirecord results, making them suitable for low-latency needs; Model Context Protocol can provide similar access for agents and is often an API wrapper. Event streams, commonly built with Kafka, Flink or Spark Streaming, suit sub-second latency and enable per-event calculations, transformations and enrichment. The excerpt also identifies Delta Change Data Feed (CDF) as a record-by-record pattern and traditional batch ETL as a multirecord pattern, but ends before explaining CDF's mechanics or comparing these approaches further.

### Source excerpt

Data is a critical component of software architecture, providing the foundation for business processes, including those powered by artificial intelligence (AI). Quality, volume, latency, consistency, ordering and duplicates are essential considerations in software design. Modern cloud architectures consume data in various ways, such as through application programming interfaces (APIs), streams and data lakes. You can consume data at the individual event or record level as it arrives or in larger batches. Legacy architectures often process data in scheduled batches, forcing surrounding architecture and business processes to depend on that approach. Getting the data consumption approach right up front is vital, because rearchitecting later is costly. General patterns to consume data There are several general data consumption patterns. These include APIs (record-by-record or multirecord results), event stream consumption (record-by-record), Delta Change Data Feed (CDF) consumption (record-by-record) or traditional batch ETL-extract, transform, load (multirecord results). APIs are ideal for low-latency, record-by-record requirements. They also support multirecord results, typically through pagination. APIs usually follow a synchronous approach, where the client makes a call and waits for a response before moving to the next step. Model Context Protocol is an industry specification for agents to consume data and can support record-by-record or batch results via pagination, similar to APIs. They are often a wrapper on top of APIs. The event stream pattern works well for sub-second latency requirements and typically runs on a tech stack like Kafka, Flink or Spark Streaming. These technologies excel at record-level calculations (like on-the-fly calculation), transformations (like column mapping) or enrichment (like adding data to an existing event). Delta Lake is often used in data lakes and serves as a transaction log to standard Apache Parquet data files (source). CDF is

## Advancing safe agentic AI: ASKS symposium highlights

DevFeed: [Advancing safe agentic AI: ASKS symposium highlights](<https://devfeed.tech/articles/advancing-safe-agentic-ai-asks-symposium-highlights-113886.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/advancing-safe-agentic-ai-asks-symposium-highlights-bbba40b19784?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-11-02T19:42:48Z

Content type: news

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [Agentic AI](<https://devfeed.tech/topics/what-is-agentic-ai.md>), [ai safety](<https://devfeed.tech/topics/ai-safety.md>), [ai security](<https://devfeed.tech/topics/ai-security.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [academic](<https://devfeed.tech/tags/academic.md>), [academic-partnerships](<https://devfeed.tech/tags/academic-partnerships.md>), [accountability](<https://devfeed.tech/tags/accountability.md>), [actions](<https://devfeed.tech/tags/actions.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [agentic-ai-systems](<https://devfeed.tech/tags/agentic-ai-systems.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-models](<https://devfeed.tech/tags/ai-models.md>), [ai-red-teaming](<https://devfeed.tech/tags/ai-red-teaming.md>), [ai-research](<https://devfeed.tech/tags/ai-research.md>), [ai-safety](<https://devfeed.tech/tags/ai-safety.md>), [ai-systems](<https://devfeed.tech/tags/ai-systems.md>), [auditing](<https://devfeed.tech/tags/auditing.md>), [autonomous](<https://devfeed.tech/tags/autonomous.md>), [code](<https://devfeed.tech/tags/code.md>), [computer-science](<https://devfeed.tech/tags/computer-science.md>), [data-science](<https://devfeed.tech/tags/data-science.md>), [highlights](<https://devfeed.tech/tags/highlights.md>), [llm](<https://devfeed.tech/tags/llm.md>), [science](<https://devfeed.tech/tags/science.md>)

### AI overview

At UIUC's 2026 ASKS Fall Symposium, researchers and industry leaders focused on how AI safety must adapt as systems move from generating text to taking multistep actions, using tools and executing code. Capital One's Prem Natarajan emphasized a destination-back research approach and argued that autonomous AI in regulated fields such as financial services requires rigorous safety mechanisms. The excerpt identifies sandboxing, auditing and accountability frameworks as priorities for industry-academic partnerships. The symposium also featured research readouts, panel debates and a hands-on AI red-teaming workshop; the available text does not provide further findings from those sessions.

### Source excerpt

Highlights from UIUC's 2026 ASKS Fall Symposium. On September 24-25, academic researchers, enterprise scientists and leaders gathered at the Thomas M. Siebel Center for Computer Science at the University of Illinois Urbana-Champaign (UIUC) for the ASKS Fall Symposium. The event marked a major milestone in the partnership between Capital One and the University of Illinois through the Capital One Center for Generative AI Safety, Knowledge Systems and Cybersecurity (ASKS). Designed to advance the safety, reliability and security of generative AI models, the symposium featured keynotes, research readouts, panel debates and a hands-on AI red-teaming workshop. From models to agents: Navigating the age of action The symposium kicked off with opening remarks and a fireside chat titled "From models to agents: rethinking AI research, safety, and cybersecurity in the age of action." Moderated by ASKS Co-Director Heng Ji, the session featured Prem Natarajan, chief scientist and chief enterprise AI officer at Capital One, alongside Nancy Amato, director of the Siebel School of Computing and Data Science. The discussion framed the fundamental transition taking place across the AI landscape: moving away from models that merely generate text to autonomous agentic AI-systems that can set goals, execute code, operate tools and take multistep actions. Natarajan highlighted Capital One's "destination-back" scientific approach, emphasizing that in highly regulated spaces like financial services, autonomous agency requires rigorous safety mechanisms. As AI stops just "talking" and starts "acting," industry-academic partnerships like ASKS become essential to build new sandboxing, auditing and accountability frameworks. Panel spotlight: Word models vs. world models One of the highlights of the day was a panel exploring the fundamental limits of current large language models: " Word models vs. world models: Do LLMs understand the world-or just describe it?" Moderated by Heng Ji, the panel b

## Capital One at EMNLP 2026 in Budapest

DevFeed: [Capital One at EMNLP 2026 in Budapest](<https://devfeed.tech/articles/capital-one-at-emnlp-2026-in-budapest-97648.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/capital-one-at-emnlp-2026-in-budapest-4f5b6a450ebc?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-10-23T13:23:19Z

Content type: news

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [AI Research](<https://devfeed.tech/topics/ai-research.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Retrieval Augmented Generation (RAG)](<https://devfeed.tech/topics/retrieval-augmented-generation-rag.md>), [LLM security](<https://devfeed.tech/topics/llm-security.md>), [ai safety](<https://devfeed.tech/topics/ai-safety.md>), [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [academic](<https://devfeed.tech/tags/academic.md>), [adaptive](<https://devfeed.tech/tags/adaptive.md>), [adversarial](<https://devfeed.tech/tags/adversarial.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-systems](<https://devfeed.tech/tags/agentic-systems.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-ml](<https://devfeed.tech/tags/ai-ml.md>), [ai-research](<https://devfeed.tech/tags/ai-research.md>), [applications](<https://devfeed.tech/tags/applications.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [attack](<https://devfeed.tech/tags/attack.md>), [authoring](<https://devfeed.tech/tags/authoring.md>), [behavior](<https://devfeed.tech/tags/behavior.md>), [california](<https://devfeed.tech/tags/california.md>), [challenges](<https://devfeed.tech/tags/challenges.md>), [collaboration](<https://devfeed.tech/tags/collaboration.md>), [complex](<https://devfeed.tech/tags/complex.md>), [conference](<https://devfeed.tech/tags/conference.md>), [consumer](<https://devfeed.tech/tags/consumer.md>), [data-science](<https://devfeed.tech/tags/data-science.md>), [detection](<https://devfeed.tech/tags/detection.md>), [naturallanguageprocessing](<https://devfeed.tech/tags/naturallanguageprocessing.md>), [science](<https://devfeed.tech/tags/science.md>)

### AI overview

The excerpt describes 16 Capital One and academic partner papers accepted at EMNLP 2026, spanning model safety, retrieval, intent resolution, evaluation and agent workflows. Examples include FinRT, which distills adaptive red-teaming into reusable adversarial generators for safety checks under fixed generation budgets; MEMGUARD, which inspects and isolates untrusted inputs before they contaminate long-term memory; and CAGE, which organizes retrieved passages as directed heterogeneous graphs to preserve coherence across chunks. SPARC-RAG combines sequential refinement with parallel exploration and context management, while AIR probes ambiguous text-to-SQL requests and asks targeted clarifying questions, even when only the schema is available. For GUI agents, HiViG adds progress summaries and visually grounded action critique; the excerpt reports relative task-success improvements of 5.8% for open-weight models and 9.0% for Gemini-3 Flash versus the strongest critics evaluated. Other highlighted work includes benchmarks for agent workflows, prompt regression, code-switching and judge bias. These are summaries of selected papers from the supplied text.

### Source excerpt

Capital One researchers and academic partners present 16 accepted papers advancing responsible AI, RAG and LLM evaluation.Advancing natural language processing through multi-sector research Natural language processing (NLP) forms the backbone of modern agentic systems, document intelligence and enterprise artificial intelligence applications. Capital One researchers, in collaboration with our academic partners, are presenting 16 accepted papers at the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP 2026), taking place in Budapest, Hungary, October 24-29. These research papers highlight ongoing investments in foundational AI/ML science, robust red-teaming, graph-encoded retrieval-augmented generation (RAG), intent resolution and trustworthy evaluation techniques. Multi-sector collaboration accelerates the transition of foundational NLP innovations into secure, reliable enterprise systems. By co-authoring several of these papers with university partners including the University of Illinois Urbana-Champaign, University of Southern California and UNC Chapel Hill, Capital One contributes to open science while addressing complex technical challenges at scale. Highlighted EMNLP 2026 research papers Papers on robustness, red-teaming and model safety: FinRT: Distilling adaptive red-teaming strategies into reusable adversarial generators in consumer finance Rikhiya Ghosh, Himanshu Kumar, Sriram Venkatapathy, Sahil Wadhwa, Alexandre Day, Andy Luo, Pranab Mohanty Large language models remain susceptible to adversarial prompts that induce policy-violating behavior in regulated financial contexts. This work introduces FinRT, a framework that distills adaptive red-teaming strategies into reusable adversarial generators, significantly improving attack detection and safety verification under fixed generation budgets. MEMGUARD: Preventing memory contamination in long-term memory-augmented large language models Hyeonjeong Ha, William Campbell, Yue Wu (academ

## USC and Capital One announce 2026-2027 AI research awards

DevFeed: [USC and Capital One announce 2026-2027 AI research awards](<https://devfeed.tech/articles/usc-and-capital-one-announce-2026-2027-ai-research-awards-87293.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/usc-and-capital-one-announce-2026-2027-ai-research-awards-f06209933d6c?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-10-14T20:20:03Z

Content type: news

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [AI Research](<https://devfeed.tech/topics/ai-research.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>), [text2sql](<https://devfeed.tech/topics/text2sql.md>), [Finance](<https://devfeed.tech/topics/finance.md>), [ai safety](<https://devfeed.tech/topics/ai-safety.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-research](<https://devfeed.tech/tags/ai-research.md>), [awards](<https://devfeed.tech/tags/awards.md>), [collaboration](<https://devfeed.tech/tags/collaboration.md>), [data-science](<https://devfeed.tech/tags/data-science.md>), [database](<https://devfeed.tech/tags/database.md>), [databases](<https://devfeed.tech/tags/databases.md>), [failure](<https://devfeed.tech/tags/failure.md>), [frameworks](<https://devfeed.tech/tags/frameworks.md>), [large-language-model](<https://devfeed.tech/tags/large-language-model.md>), [llm](<https://devfeed.tech/tags/llm.md>), [localization](<https://devfeed.tech/tags/localization.md>), [monte-carlo](<https://devfeed.tech/tags/monte-carlo.md>), [multi-agent](<https://devfeed.tech/tags/multi-agent.md>), [research](<https://devfeed.tech/tags/research.md>), [schema](<https://devfeed.tech/tags/schema.md>), [science](<https://devfeed.tech/tags/science.md>), [sql](<https://devfeed.tech/tags/sql.md>), [uncertainty-quantification](<https://devfeed.tech/tags/uncertainty-quantification.md>), [workflows](<https://devfeed.tech/tags/workflows.md>)

### AI overview

Capital One and USC's CREDIF selected six 2026-2027 research collaborations pairing faculty and PhD fellows with Capital One scientists, plus two inaugural faculty fellowships. Projects address uncertainty and failure localization in multi-agent LLMs; schema-guided Monte Carlo for text-to-SQL; grounded financial-table question answering; explainable financial analyst agents; graph-based financial knowledge extraction; and reward modeling for trustworthy AI. The fellowships provide two years of research support and collaboration with Capital One scientists. Projects run during the academic year, with results intended for peer-reviewed conference papers, open publications and code, and graduate-student career opportunities.

### Source excerpt

Capital One and USC's Center for Responsible AI and Decision Making in Finance (CREDIF) celebrate 2026-2027 AI research awardees.Advancing responsible AI through multi-sector academic collaboration Capital One selected the 2026-2027 CREDIF Research Award Recipients for joint academic collaborations with the University of Southern California (USC). Established through the Center for Responsible AI and Decision Making in Finance (CREDIF) at USC, these collaborations pair academic faculty with Capital One research scientists to address foundational challenges in machine learning and financial AI systems. Along with these award recipients, two faculty members were chosen to receive our inaugural CREDIF Capital One Faculty Fellowship Award. Granted as a two-year gift to professors spearheading pioneering AI research, these fellowships provide direct support for ongoing research efforts while enabling recipients and their graduate students to actively collaborate with Capital One scientists. The 2026-2027 awarded research collaborations and principal investigators include: Uncertainty quantification for multi-agent LLM systems: Decomposition and failure localization Faculty Lead: Prof. Salman Avestimehr PhD Fellow: Asal Mehradfar Focus: Developing robust frameworks to quantify uncertainty, decompose complex agent workflows and localize failure modes in multi-agent large language model environments. Dynamic schema-guided Monte Carlo for text-to-SQL agents on massive databases Faculty Lead: Prof. Paul Bogdan PhD Fellow: Kien Nguyen Focus: Advancing natural language interfaces for database querying by pairing dynamic schema representation with Monte Carlo search techniques. FinSemKG: A semantic mediation layer for grounded, auditable LLM question answering over financial tables Faculty Lead: Prof. Craig Knoblock PhD Fellow: Md Abrar Jahin Focus: Constructing semantic knowledge graph structures that ensure grounded, explainable and auditable question-answering pipelines for t

## Operational excellence in serverless -- 5E's framework

DevFeed: [Operational excellence in serverless -- 5E's framework](<https://devfeed.tech/articles/operational-excellence-in-serverless-5e-s-framework-84374.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/operational-excellence-in-serverless-5es-framework-f3bf3dcd2cda?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-10-09T15:27:13Z

Content type: article

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [Serverless](<https://devfeed.tech/topics/serverless.md>), [Production Engineering](<https://devfeed.tech/topics/production-engineering.md>), [serverless monitoring](<https://devfeed.tech/topics/serverless-monitoring.md>), [engineering-culture](<https://devfeed.tech/topics/engineering-culture.md>), [cloud cost management](<https://devfeed.tech/topics/cloud-cost-management.md>), [CI/CD](<https://devfeed.tech/topics/cicd.md>), [telemetry](<https://devfeed.tech/topics/telemetry.md>), [DevOps](<https://devfeed.tech/topics/devops.md>), [policy-as-code](<https://devfeed.tech/topics/policy-as-code.md>), [Chaos Engineering](<https://devfeed.tech/topics/chaos-engineering.md>), [AWS Step Functions](<https://devfeed.tech/topics/aws-step-functions.md>)

Tags: [agile](<https://devfeed.tech/tags/agile.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [automation](<https://devfeed.tech/tags/automation.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [cloud-computing](<https://devfeed.tech/tags/cloud-computing.md>), [continuous](<https://devfeed.tech/tags/continuous.md>), [continuous-improvement](<https://devfeed.tech/tags/continuous-improvement.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cost-control](<https://devfeed.tech/tags/cost-control.md>), [culture](<https://devfeed.tech/tags/culture.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [distributed-workloads](<https://devfeed.tech/tags/distributed-workloads.md>), [framework](<https://devfeed.tech/tags/framework.md>), [observability](<https://devfeed.tech/tags/observability.md>), [operational](<https://devfeed.tech/tags/operational.md>), [operational-excellence](<https://devfeed.tech/tags/operational-excellence.md>), [reliability](<https://devfeed.tech/tags/reliability.md>), [resilience](<https://devfeed.tech/tags/resilience.md>), [server](<https://devfeed.tech/tags/server.md>), [serverless](<https://devfeed.tech/tags/serverless.md>), [software-architecture](<https://devfeed.tech/tags/software-architecture.md>), [software-engineering](<https://devfeed.tech/tags/software-engineering.md>), [software-systems](<https://devfeed.tech/tags/software-systems.md>)

### AI overview

The 5E framework treats serverless operational excellence as ongoing engineering and team practice: empower teams with infrastructure as code, CI/CD, policy checks and self-service deployment; engineer for reliability with idempotent functions, dead-letter queues, orchestration, asynchronous eventing and end-to-end observability; evaluate performance and cost using shared metrics; evolve configurations and architectures based on usage data; and enhance through blameless postmortems, alerts and measurable retrospectives. Serverless reduces infrastructure management but distributed, event-driven workloads still require deliberate resilience, cost visibility and continuous learning. The supplied text gives concrete practices, but its Evaluate section names no specific metrics beyond general performance and cost signals.

### Source excerpt

Transform serverless freedom into business agility through the 5E's framework: empower, engineer, evaluate, evolve and enhance. Modern digital businesses need software systems that are agile, reliable and cost-efficient. At Capital One, we've adopted a serverless-first approach to transform how we build and deploy applications, trading server management for a focus on delivering business value. While this paradigm eliminates operational overhead related to infrastructure, it doesn't eliminate the need for operational excellence. In fact, the distributed, event-driven nature of serverless systems introduces a new set of challenges. When looking to achieve operational excellence in serverless systems, organizations must require disciplined engineering-automation, observability, cost control and a culture of continuous improvement. Understanding operational excellence in the serverless context Operational excellence focuses on how cloud systems are designed, deployed, monitored and optimized for long-term performance and reliability. In a serverless context, it involves: Automation: infrastructure and deployment as code Resilience: designing for failures, timeouts and throttling from day one Observability: collecting logs, metrics and traces across functions and services Cost visibility: paying only for what's used, while avoiding hidden inefficiencies Continuous improvement: using data and feedback to refine architecture Serverless architecture can reduce reliance on infrastructure management, but it necessitates robust operational discipline to effectively handle distributed workloads. We will now delve into the five guiding principles, known as the 5E's, which bridge the gap between operational excellence and practical engineering practices. Figure 1: Operational excellence illustrative architecture -- The 5E's Framework for Resilience & AgilityEmpower: automation and governance Empowerment starts with automation and clarity. Teams must have the freedom to deploy qui

## Announcing Capital One Fellows for academic year 2026-2027

DevFeed: [Announcing Capital One Fellows for academic year 2026-2027](<https://devfeed.tech/articles/announcing-capital-one-fellows-for-academic-year-2026-2027-64237.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/announcing-capital-one-fellows-for-academic-year-2026-2027-b1fc18e45b6f?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-10-02T15:55:59Z

Content type: news

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [AI Engineering](<https://devfeed.tech/topics/ai-engineering.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-machine-learning](<https://devfeed.tech/tags/ai-machine-learning.md>), [ai-research](<https://devfeed.tech/tags/ai-research.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [data-science](<https://devfeed.tech/tags/data-science.md>), [developers](<https://devfeed.tech/tags/developers.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [science](<https://devfeed.tech/tags/science.md>)

### AI overview

Capital One introduces its 2026-2027 doctoral fellows and their research projects in AI, machine learning, and data science. The projects include code language models and software security, human-AI collaboration for work that includes software development, and research into reliable AI systems.

### Source excerpt

Meet the 2026-2027 Capital One Fellows advancing doctoral research in artificial intelligence, machine learning and data science. We're thrilled to introduce the Capital One Fellows award recipients for the 2026-2027 academic year. The Capital One Fellows program is an annual fellowship for doctoral students, designed to equip emerging researchers with the skills and knowledge needed to excel in the field of AI, machine learning and data science. This exceptional group of researchers is working alongside leading faculty to tackle complex challenges and push the boundaries of their field. (Learn more about our academic partnerships.) Here's a closer look at this year's Capital One Fellows and their upcoming projects. Columbia UniversityTao Long Project title: Vibe-Working: Agentic Infrastructure for Humans to Become AI Supervisors for General Knowledge Work Faculty advisor: Dr. Lydia Chilton Tao Long is a computer science PhD student at Columbia University, advised by Professor Lydia Chilton. Addressing human-AI interaction, Tao's research explores how humans collaborate with generative AI systems and AI agents over time, focusing on making AI tools more usable, useful, trustworthy, reliable and seamlessly integrated into everyday productivity practices. Specifically, Tao builds human-AI and agentic systems that reduce cognitive and temporal effort for challenging or complex tasks; offload work to AI while maintaining human ownership and authenticity; and fit naturally into the existing processes of writers, developers, designers, event organizers and many other communities. Before starting their PhD, Tao earned a BS summa cum laude from Cornell University. Jinjun Peng Project title: Secure CodeLLM with Reasoning Faculty advisor: Dr. Baishakhi Ray Jinjun Peng is a PhD student in computer science at Columbia University, advised by Professors Baishakhi Ray and Junfeng Yang. His research has centered on code language models, including SWE-Spot, which builds small reposi

## Beyond the hallucination: Why AI needs deterministic math

DevFeed: [Beyond the hallucination: Why AI needs deterministic math](<https://devfeed.tech/articles/beyond-the-hallucination-why-ai-needs-deterministic-math-63603.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/beyond-the-hallucination-why-ai-needs-deterministic-math-aa9c448479ab?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-10-01T19:31:24Z

Content type: article

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [AI Development](<https://devfeed.tech/topics/ai-development.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Occam's razor machine learning](<https://devfeed.tech/topics/occam-s-razor-machine-learning.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [aifluency](<https://devfeed.tech/tags/aifluency.md>), [artificial-intelligence-machine-learning](<https://devfeed.tech/tags/artificial-intelligence-machine-learning.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [compliance](<https://devfeed.tech/tags/compliance.md>), [costs](<https://devfeed.tech/tags/costs.md>), [critical](<https://devfeed.tech/tags/critical.md>), [generative-ai-tools](<https://devfeed.tech/tags/generative-ai-tools.md>)

### AI overview

The article recommends combining probabilistic AI for discovery, prediction, and categorization with deterministic code for precise calculations, aggregation, and validation. Its example classifies invoice transactions with AI, sums costs in a deterministic process, then uses AI to narrate the results. It argues that this separation improves auditability, reliability, and problem isolation for operational metrics and compliance.

### Source excerpt

AI excels at discovery and pattern recognition, but financial math demands 100% verifiable, deterministic calculation.Why core operational metrics demand more than probabilistic AI While Artificial Intelligence/Machine Learning models excel at pattern recognition, anomaly detection and predictive analytics, relying on them alone for precise data aggregation can potentially lead to inconsistencies and a loss of trust in your data. We've all heard about "AI hallucinations" in LLMs-imagine that margin of error applied to your core operational metrics. Most AI/ML models are inherently probabilistic: They identify patterns based on likelihoods, not certainties. For business-critical functions, such as aggregating cloud costs or ensuring regulatory compliance, "highly likely" isn't good enough. There's a need for verifiable accuracy. This isn't an argument against AI use for operational insights-quite the opposite. The real power lies in a hybrid approach that uses AI for what it does best (discovery, prediction, identification of opportunities) and couples it with deterministic implementations for what must be precise (calculation, aggregation, validation). Think of it as building an insight engine with two distinct but integrated components: The AI/ML discovery layer sifts through vast datasets to identify anomalies, forecast trends, categorize unstructured system logs, and surface optimization opportunities. It answers "what to look at" and "why it might be happening." The deterministic calculation layer takes the AI's outputs and executes precise, rule-based, verifiable computations. This is where performance metrics are summed and compliance checks run with accuracy. This architecture ensures that while we benefit from AI surfacing complex insights, the underlying core business metrics remain unassailable. Implementing deterministic wrappers The key to operationalizing this hybrid approach is designing systems with clear boundaries so that any data requiring absolute

## CAIRFI and Capital One announce 2026-2027 AI Research Awards

DevFeed: [CAIRFI and Capital One announce 2026-2027 AI Research Awards](<https://devfeed.tech/articles/cairfi-and-capital-one-announce-2026-2027-ai-research-awards-62432.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/cairfi-and-capital-one-announce-2026-2027-ai-research-awards-46178d80e007?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-09-30T13:14:41Z

Content type: news

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Uncertainty quantification LLMs](<https://devfeed.tech/topics/uncertainty-quantification-llms.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [LLM security](<https://devfeed.tech/topics/llm-security.md>), [ai safety](<https://devfeed.tech/topics/ai-safety.md>), [post-training](<https://devfeed.tech/topics/post-training.md>), [ai security](<https://devfeed.tech/topics/ai-security.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [academic](<https://devfeed.tech/tags/academic.md>), [academic-research](<https://devfeed.tech/tags/academic-research.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-research](<https://devfeed.tech/tags/ai-research.md>), [ai-security](<https://devfeed.tech/tags/ai-security.md>), [announce](<https://devfeed.tech/tags/announce.md>), [apis](<https://devfeed.tech/tags/apis.md>), [application-programming-interface](<https://devfeed.tech/tags/application-programming-interface.md>), [award](<https://devfeed.tech/tags/award.md>), [awards](<https://devfeed.tech/tags/awards.md>), [causal-inference](<https://devfeed.tech/tags/causal-inference.md>), [causal-machine-learning](<https://devfeed.tech/tags/causal-machine-learning.md>), [columbia-university](<https://devfeed.tech/tags/columbia-university.md>), [continual-learning](<https://devfeed.tech/tags/continual-learning.md>), [data-science](<https://devfeed.tech/tags/data-science.md>), [financial](<https://devfeed.tech/tags/financial.md>), [financial-data](<https://devfeed.tech/tags/financial-data.md>), [financial-innovation](<https://devfeed.tech/tags/financial-innovation.md>), [financial-institutions](<https://devfeed.tech/tags/financial-institutions.md>), [research](<https://devfeed.tech/tags/research.md>), [science](<https://devfeed.tech/tags/science.md>)

### AI overview

CAIRFI and Capital One awarded support for 2026-2027 faculty research on responsible AI in financial services. The excerpt lists five projects: Micah Goldblum is developing uncertainty estimates for black-box LLM APIs to help assess output reliability; Adam Block is studying post-training reinforcement learning dynamics to improve model stability; Kathleen McKeown is working on fast, controllable update summarization that preserves source fidelity; Lydia Chilton is exploring agentic systems for regulatory compliance and policy navigation; and Baishaki Ray is integrating reasoning into code-generation LLMs to identify and mitigate security vulnerabilities. The excerpt also names Junfeng Yang's project on proactive defenses against LLM jailbreaks but provides no further details.

### Source excerpt

The Columbia Center for AI and Responsible Financial Innovation (CAIRFI) and Capital One are proud to announce the recipients of the 2026-2027 AI research awards. Now in its third year, this collaboration continues to build on the mission of advancing responsible AI and innovation in financial services. By combining Columbia University's world-class academic research with Capital One's industry-leading expertise, CAIRFI aims to solve complex challenges in financial services through cutting-edge AI. This year, the center issued a Call for Proposals seeking faculty AI research focused on several key pillars of modern financial technology: Continual learning Reinforcement learning Guardrails and jailbreaking in large language models Responsible and trustworthy AI; AI security Causal inference and causal machine learning Multi-agentic LLMs for complex reasoning Reasoning in sequential decision settings Multimodality beyond vision and language: LLMs + tabular, LLMs + time series Generative AI for synthetic financial data Differential privacy, and its applications to information transmission Other innovative methods or ideas AI Research Award recipients of the 2026 to 2027 academic year Following a rigorous selection process, the following faculty and research projects have been awarded support for the upcoming academic year. Micah Goldblum Uncertainty Estimates for Black-Box Application Programming Interface Language Models As financial institutions increasingly rely on third-party APIs for large language model (LLM) capabilities, understanding when to trust a model's output is critical. Professor Goldblum's research focuses on developing robust methods to estimate uncertainty in "black box" models. This work ensures that AI-driven financial insights are backed by measurable reliability, reducing the risk of overreliance on incorrect model outputs. Adam Block Understanding and Stabilizing Reinforcement Learning Dynamics in Post-Training Language Models Reinforcement lear

## Strengthening open source security with Socket

DevFeed: [Strengthening open source security with Socket](<https://devfeed.tech/articles/strengthening-open-source-security-with-socket-59587.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/strengthening-open-source-security-with-socket-7ee8fe88bfec?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-09-24T16:36:07Z

Content type: article

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [open-source-security](<https://devfeed.tech/topics/open-source-security.md>), [software composition analysis](<https://devfeed.tech/topics/software-composition-analysis.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>)

Tags: [cloud-security](<https://devfeed.tech/tags/cloud-security.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [security](<https://devfeed.tech/tags/security.md>), [socket](<https://devfeed.tech/tags/socket.md>), [software-engineering](<https://devfeed.tech/tags/software-engineering.md>), [software-supply-chain-security](<https://devfeed.tech/tags/software-supply-chain-security.md>), [tech-startups](<https://devfeed.tech/tags/tech-startups.md>), [venture-capital](<https://devfeed.tech/tags/venture-capital.md>)

### AI overview

Capital One uses Socket to analyze open source packages for indicators of risky behavior before they enter its environment. This adds an earlier signal to traditional approaches that primarily identify previously disclosed vulnerabilities, helping security and engineering teams assess dependencies with more timely context. The stated capabilities include detecting potentially malicious or compromised packages, improving dependency visibility and hygiene, and supporting developer workflows. Capital One Ventures participated in Socket's $60 million Series C round, citing the growing importance of supply chain security amid faster software development and AI-powered coding.

### Source excerpt

Delivering real-time dependency visibility and empowering secure developer workflows at scale. Open source software is foundational to modern application development. It helps engineering teams build faster, adopt new capabilities and benefit from the work of a global developer community. But as the use of open source grows, so does the importance of understanding and managing the associated security risks. At Capital One, software supply chain security is part of our broader commitment to collective defense. Our goal is to give developers the tools and context they need to use open source confidently while maintaining strong security standards. Sharpening visibility into open-source risk Modern software supply chains are deeply interconnected. A single vulnerability in a widely used open source component can ripple across thousands of enterprises simultaneously. Capital One uses Socket as part of its broader, proactive approach to open source and software supply chain security. Socket analyzes open source packages for indicators of risky behavior before those packages can enter the environment. This complements traditional approaches that primarily identify previously disclosed vulnerabilities. By surfacing relevant information earlier in the development life cycle, Socket helps security and engineering teams make more-informed decisions about the software dependencies they use. Socket is focused on strengthening the ability to: Identify potentially malicious or compromised packages before they enter the environment Improve visibility into open source dependencies Promote stronger dependency hygiene Give developers timely and actionable context These capabilities support a more proactive approach to software supply chain security while allowing developers to continue benefiting from the speed and flexibility of open source. Why Capital One Ventures invested in Socket Capital One Ventures, our strategic investing arm, recently participated in Socket's $60 million Se

## Scaling Enterprise Testing on Kubernetes with Argo

DevFeed: [Scaling Enterprise Testing on Kubernetes with Argo](<https://devfeed.tech/articles/scaling-enterprise-testing-on-kubernetes-with-argo-42802.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/scaling-enterprise-testing-on-kubernetes-with-argo-5e51f3df3ebc?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-09-18T14:01:33Z

Content type: article

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [Software Testing](<https://devfeed.tech/topics/software-testing.md>), [Kubernetes](<https://devfeed.tech/topics/kubernetes.md>), [Scalability](<https://devfeed.tech/topics/scalability.md>), [Orchestration](<https://devfeed.tech/topics/orchestration.md>), [Multi-tenancy](<https://devfeed.tech/topics/multi-tenancy.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>), [Cloud Native Ecosystem](<https://devfeed.tech/topics/cloud-native-ecosystem.md>), [Parallelism](<https://devfeed.tech/topics/parallelism.md>), [Latency](<https://devfeed.tech/topics/latency.md>)

Tags: [argo](<https://devfeed.tech/tags/argo.md>), [ci-cd](<https://devfeed.tech/tags/ci-cd.md>), [cloud-native](<https://devfeed.tech/tags/cloud-native.md>), [cluster](<https://devfeed.tech/tags/cluster.md>), [cncf](<https://devfeed.tech/tags/cncf.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [kubernetes](<https://devfeed.tech/tags/kubernetes.md>), [latency](<https://devfeed.tech/tags/latency.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [parallel](<https://devfeed.tech/tags/parallel.md>), [scaling](<https://devfeed.tech/tags/scaling.md>), [software-engineering](<https://devfeed.tech/tags/software-engineering.md>), [testing](<https://devfeed.tech/tags/testing.md>)

### AI overview

A case study of Capital One's rebuilt enterprise testing platform, which uses Kubernetes and open-source Argo ecosystem projects to run large numbers of parallel tests across teams. It describes the limitations of the previous pipeline and the platform's focus on scalability, multi-tenancy, and stability.

### Source excerpt

How we built a resilient, multi-tenant test platform on open-source workflows. Every engineer knows the quiet anxiety of watching a CI/CD pipeline run. You commit, you wait for the build and then you wait some more. Now imagine that at enterprise scale: thousands of test agents running at once, across dozens of teams, all validating business-critical services on the same shared infrastructure. That's the problem our platform exists to solve, and for years, the way we solved it was quietly holding us back. At Capital One, we're a technology company that happens to operate in one of the most tightly regulated industries there is. That combination raises the bar: our internal platforms strive to be fast and safe, to scale for many teams at once and to prevent any single workload from destabilizing the rest. So when our first-generation testing pipeline started hitting hard limits, we didn't just move it; we rebuilt it on Kubernetes using open-source, Cloud Native Computing Foundation (CNCF) projects from the Argo ecosystem. In this post, we'll walk through three things: Why we moved off our first-generation pipeline How Argo orchestrates a single test run, end to end The guardrails that keep the platform stable when thousands of tests arrive at once The headline isn't just "we moved to Kubernetes." It's that we learned to run massively parallel test workloads without letting any single burst threaten a shared, multi-tenant cluster. Why we moved off our first-generation pipeline Our original platform got the job done, but it was built from a highly complex set of managed services linked together by a proprietary orchestrator. Execution logic lived inside cloud-specific state-machine definitions, far away from the containers actually running the tests. Three pain points stood out: State-machine sprawl. Changing execution logic meant editing large, cloud-specific workflow definitions. The orchestration lived far away from the code it was orchestrating. Cold-start latency.

## Advancing AI fluency at Capital One

DevFeed: [Advancing AI fluency at Capital One](<https://devfeed.tech/articles/advancing-ai-fluency-at-capital-one-22568.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/advancing-ai-fluency-at-capital-one-2388f75e0139?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-08-11T15:12:03Z

Content type: article

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Learning](<https://devfeed.tech/topics/learning.md>), [software-development](<https://devfeed.tech/topics/software-development.md>), [agentic-coding](<https://devfeed.tech/topics/agentic-coding.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>)

Tags: [advanced-ai-capabilities](<https://devfeed.tech/tags/advanced-ai-capabilities.md>), [advancing-ai](<https://devfeed.tech/tags/advancing-ai.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-coding-tools](<https://devfeed.tech/tags/ai-coding-tools.md>), [ai-learning](<https://devfeed.tech/tags/ai-learning.md>), [ai-literacy](<https://devfeed.tech/tags/ai-literacy.md>), [ai-machine-learning](<https://devfeed.tech/tags/ai-machine-learning.md>), [aifluency](<https://devfeed.tech/tags/aifluency.md>), [architect](<https://devfeed.tech/tags/architect.md>), [bank](<https://devfeed.tech/tags/bank.md>), [build](<https://devfeed.tech/tags/build.md>), [business](<https://devfeed.tech/tags/business.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [cloud-engineering](<https://devfeed.tech/tags/cloud-engineering.md>), [coding](<https://devfeed.tech/tags/coding.md>), [coding-tools](<https://devfeed.tech/tags/coding-tools.md>), [college](<https://devfeed.tech/tags/college.md>), [company](<https://devfeed.tech/tags/company.md>), [complex](<https://devfeed.tech/tags/complex.md>), [content](<https://devfeed.tech/tags/content.md>), [courses](<https://devfeed.tech/tags/courses.md>)

### AI overview

Capital One describes AI fluency as a company-wide operating capability built on its earlier investments in data, cloud and in-house engineering. It launched an AI learning hub for more than 60,000 associates, offering paths from foundational literacy to advanced workflows and domain-specific tools. For engineers, AI coding tools are intended to streamline routine tasks and free attention for architecture and complex problem-solving. The company says learning is built into work through dedicated experimentation time, leadership support and hands-on training with Google and Anthropic. A month-long agentic coding program involved more than 10,000 engineers in a live session, followed by weekly workshops and coding challenges based on enterprise problems; the curriculum is expanding to agentic developer experience, multi-agent orchestration and token efficiency.

### Source excerpt

Capital One is raising the bar on AI fluency by making it a fundamental part of how we operate. At Capital One, we sought early on to operate like a bank that a technology company would build--modernizing our data ecosystem, going all in on the cloud and building an in-house engineering workforce. That deep investment has set us up to thrive with AI at enterprise scale today. As we continue to build at the forefront of technology and AI, sustaining an edge requires more than just deploying off-the-shelf models. Because at Capital One, we don't just use AI--we build AI. Delivering proprietary AI solutions across the enterprise requires investing deeply in the people who architect, customize and build with them. That's why our enterprise-wide emphasis on AI fluency isn't an ad-hoc training program. It is a fundamental part of how we operate, innovate and push the boundaries of what's possible for our business and our more than 100 million customers. AI skill building at enterprise scale Building on nearly a decade of insights from internal learning platforms and initiatives like Tech College--which offers hundreds of courses and offerings on topics like AI, machine learning, data, cloud engineering and beyond--this year we launched a dynamic, enterprise-wide AI learning hub to over 60,000 Capital One associates. More than a static content library, the platform meets associates where they are, whether that means building foundational AI literacy, mastering advanced workflows or getting training for specialized tools purpose-fit for their domains. We believe AI fluency shouldn't be treated as a side-of-desk exercise or an isolated science project; it must be a shared capability across the entire company. This is especially true for our engineers and developers. AI coding tools have redefined what high-performance software development looks like. By leveraging advanced AI capabilities to help streamline "run-the-engine" tasks, our engineers are more free to focus on complex

## Highlights from MLSys 2026

DevFeed: [Highlights from MLSys 2026](<https://devfeed.tech/articles/highlights-from-mlsys-2026-22573.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/highlights-from-mlsys-2026-5e6d9f226f3d?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-08-11T15:07:08Z

Content type: news

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Retrieval Augmented Generation (RAG)](<https://devfeed.tech/topics/retrieval-augmented-generation-rag.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [LLM Techniques](<https://devfeed.tech/topics/llm-techniques.md>), [AI Architecture](<https://devfeed.tech/topics/ai-architecture.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-research](<https://devfeed.tech/tags/ai-research.md>), [attention](<https://devfeed.tech/tags/attention.md>), [cache](<https://devfeed.tech/tags/cache.md>), [communication](<https://devfeed.tech/tags/communication.md>), [compute](<https://devfeed.tech/tags/compute.md>), [computer-vision](<https://devfeed.tech/tags/computer-vision.md>), [conference](<https://devfeed.tech/tags/conference.md>), [faster](<https://devfeed.tech/tags/faster.md>), [generative](<https://devfeed.tech/tags/generative.md>), [generative-ai](<https://devfeed.tech/tags/generative-ai.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [highlights](<https://devfeed.tech/tags/highlights.md>), [inference](<https://devfeed.tech/tags/inference.md>), [inference-optimization](<https://devfeed.tech/tags/inference-optimization.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [insights](<https://devfeed.tech/tags/insights.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [science](<https://devfeed.tech/tags/science.md>)

### AI overview

Capital One's MLSys 2026 recap highlights systems research aimed at making LLM and agent workloads more efficient. For LLM inference, it points to work reducing self-attention costs and overlapping computation with communication. For latency-constrained RAG, it notes vector-index storage and recall optimization, rank-aware early termination, and prefetching retrieved context during generation. For agentic systems, it highlights planning and memory optimizations and multi-agent approaches to improving training and serving; the recap observes that mature agent workloads can benefit from co-designing agents and compute stacks. The article also describes Capital One associates presenting research and networking with students and industry professionals.

### Source excerpt

Capital One's AI research team recaps MLSys 2026, including optimizing serving LLMs, RAG and agentic AI. The 9th Annual Conference on Machine Learning and Systems (MLSys) took place in May in Bellevue, Washington. MLSys is a highly selective interdisciplinary conference sitting at the intersection of machine learning (ML) and systems design. The conference highlights cutting-edge research that combines generative AI, natural language processing, computer vision and reinforcement learning with infrastructure, deployment and hardware optimizations to make AI faster, scalable and more performant. MLSys offered Capital One associates the opportunity to learn from world-class conference sessions presented by experts in the field. All the attending associates left brimming with new ideas and planned collaborations. Kel Vanee, MVP, Machine Learning Engineering, presented some of the work happening at Capital One on using AI to make AI more efficient. Takeaways and favorite papers from MLSys 2026 Some of the most prevalent topics at MLSys this year were on cache management, model speculation, retrieval augmented generation (RAG) and agentic AI. With a plethora of relevant and interesting talks, we had no shortage of papers to choose favorites from. While a complete list of the papers we loved would be far too long, here are a few standouts: Large language model inference optimization One of the leading themes this year was how to more efficiently serve LLM models. We especially liked the papers on reducing self-attention costs, such as MAC-Attention: a Match-Amend-Complete scheme for fast and accurate attention computation and BLASST: Dynamic BLocked Attention Sparsity via Softmax Thresholding. We found valuable insights in papers covering how best to overlap computation with communication, such as TokenWeave: Efficient Compute-Communication Overlap for Distributed LLM Inference, Stream2LLM: Overlap Context Streaming and Prefill for Reduced Time-to-First-Token and FlashAgen

## Introducing AI-powered container standardization

DevFeed: [Introducing AI-powered container standardization](<https://devfeed.tech/articles/introducing-ai-powered-container-standardization-22574.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/introducing-ai-powered-container-standardization-2f9314cde883?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-07-27T15:04:11Z

Content type: article

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [container image compliance](<https://devfeed.tech/topics/container-image-compliance.md>), [container](<https://devfeed.tech/topics/container.md>), [code security](<https://devfeed.tech/topics/code-security.md>), [software composition analysis](<https://devfeed.tech/topics/software-composition-analysis.md>), [CI/CD](<https://devfeed.tech/topics/cicd.md>), [Test automation](<https://devfeed.tech/topics/test-automation.md>), [AI software supply chain security](<https://devfeed.tech/topics/ai-software-supply-chain-security.md>), [distroless](<https://devfeed.tech/topics/distroless.md>), [container native](<https://devfeed.tech/topics/container-native.md>), [Continuous Delivery (CD)](<https://devfeed.tech/topics/continuous-delivery.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-powered](<https://devfeed.tech/tags/ai-powered.md>), [ai-security](<https://devfeed.tech/tags/ai-security.md>), [approach](<https://devfeed.tech/tags/approach.md>), [artificial-intelligence-ai](<https://devfeed.tech/tags/artificial-intelligence-ai.md>), [attack-surface](<https://devfeed.tech/tags/attack-surface.md>), [automate](<https://devfeed.tech/tags/automate.md>), [automatically](<https://devfeed.tech/tags/automatically.md>), [avoid](<https://devfeed.tech/tags/avoid.md>), [base-images](<https://devfeed.tech/tags/base-images.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [compliance](<https://devfeed.tech/tags/compliance.md>), [container](<https://devfeed.tech/tags/container.md>), [container-security](<https://devfeed.tech/tags/container-security.md>), [containers](<https://devfeed.tech/tags/containers.md>), [delivery](<https://devfeed.tech/tags/delivery.md>), [dependencies](<https://devfeed.tech/tags/dependencies.md>), [detection](<https://devfeed.tech/tags/detection.md>), [developers](<https://devfeed.tech/tags/developers.md>), [development](<https://devfeed.tech/tags/development.md>), [software-engineering](<https://devfeed.tech/tags/software-engineering.md>), [standardization](<https://devfeed.tech/tags/standardization.md>)

### AI overview

The article describes a container security program built around preventing vulnerable components from entering the software supply chain and automating fixes for issues that do emerge. It recommends curated, hardened base images, vetted dependencies, lean production images, and SBOM checks to identify risky or abandoned libraries. The platform continuously scans source code, dependencies, container images, and runtime workloads; AI assesses contextual risk, including code reachability, data sensitivity, and threat intelligence, to prioritize remediation. When updates are available, orchestration generates pull requests, runs automated tests, and promotes patches through progressive rollouts, with scanning and audit evidence built into governance. The article stresses that safe automation depends on test maturity: well-tested applications may be deployed automatically, while others may need human review; canaries, blue-green releases, and rollback help limit deployment issues. It reports MTTR falling from weeks to hours, while presenting these as the organization's results rather than guaranteed outcomes.

### Source excerpt

AI-powered container standardization enables hands-free security remediation without sacrificing development velocity. In today's cloud-native world, containers have become the foundation of modern software delivery. Managing container security at scale can present significant challenges, thousands of vulnerabilities to track, manual remediation processes, inconsistent base images across teams and compliance requirements that slow everything down. AI-powered container standardization is a comprehensive approach that leverages artificial intelligence (AI) to automate vulnerability detection, prioritization and remediation across your entire container ecosystem. This transformative solution significantly reduces the manual toil traditionally associated with keeping containers secure and compliant. Development teams can focus on building features while the system automatically identifies vulnerabilities, generates fixes, tests changes and deploys patches, minimizing human intervention for routine updates. This approach simplifies the coordination of vulnerability remediation across hundreds of teams while ensuring that security and compliance are embedded throughout the software life cycle. Organizations can empower their developers to focus on innovation rather than patching, resulting in faster releases, stronger security and more productive engineering teams. The strategic framework: avoid and automate Our strategy for managing vulnerabilities focuses on two core principles: avoiding exposure and automating remediation, with governance baked into the software development life cycle. Avoid: Secure the foundation to minimize risk entering your environment. The most effective vulnerability remediation happens prior to the vulnerability reaching your digital ecosystems. By controlling what enters the software supply chain through curated base images and vetted dependencies/packages, you dramatically reduce the attack surface before code ever runs in production. This inc

## Announcing Capital One's 2026 UIUC AI Awardees

DevFeed: [Announcing Capital One's 2026 UIUC AI Awardees](<https://devfeed.tech/articles/announcing-capital-one-s-2026-uiuc-ai-awardees-22569.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/announcing-capital-ones-2026-uiuc-ai-awardees-729fc61a899d?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-07-21T14:45:44Z

Content type: news

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Hallucination detection](<https://devfeed.tech/topics/hallucination-detection.md>), [ai safety](<https://devfeed.tech/topics/ai-safety.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [2026](<https://devfeed.tech/tags/2026.md>), [academic](<https://devfeed.tech/tags/academic.md>), [academic-research](<https://devfeed.tech/tags/academic-research.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-research](<https://devfeed.tech/tags/ai-research.md>), [ai-safety](<https://devfeed.tech/tags/ai-safety.md>), [align](<https://devfeed.tech/tags/align.md>), [also](<https://devfeed.tech/tags/also.md>), [announce](<https://devfeed.tech/tags/announce.md>), [applications](<https://devfeed.tech/tags/applications.md>), [based](<https://devfeed.tech/tags/based.md>), [bridging](<https://devfeed.tech/tags/bridging.md>), [but](<https://devfeed.tech/tags/but.md>), [code](<https://devfeed.tech/tags/code.md>), [code-generation](<https://devfeed.tech/tags/code-generation.md>), [component](<https://devfeed.tech/tags/component.md>), [core](<https://devfeed.tech/tags/core.md>), [research](<https://devfeed.tech/tags/research.md>), [science](<https://devfeed.tech/tags/science.md>), [uiuc](<https://devfeed.tech/tags/uiuc.md>)

### AI overview

The excerpt describes three 2026-2027 University of Illinois projects supported by Capital One's ASKS partnership. Hao Peng's work seeks to improve LLM reasoning faithfulness without traditional reward systems, which the source characterizes as often biased. Talia Ringer and Reyhan Jabaarvand's TrACE-Contrast applies trace-aware contrastive learning to align code generation with execution traces. Heng Ji's project explores imaginative LLM agents that consider multiple possible outcomes before acting. The excerpt frames these approaches as contributions to trustworthy reasoning, code integrity, and creative problem-solving; it does not provide details for all five projects.

### Source excerpt

Meet the University of Illinois researchers and fellows advancing Agentic AI through our academic partnership.Announcing the Center for Generative AI Safety, Knowledge Systems, and Cybersecurity (ASKS) 2026-2027 Capital One Research Awardees from the University of Illinois As we look toward the 2026-2027 academic year, we continue to partner with institutions that lead the global conversation on the future of intelligence. We are thrilled to announce this year's cohort of research and fellowship awardees from the University of Illinois, whose pioneering work addresses the most critical frontier in technology today: Agentic AI. From the way machines "think" and "imagine" to the hardware that powers them and the safety protocols that govern them, these five projects represent Capital One's holistic push toward AI that is not only powerful but also faithful, safe and creative. The 2026-2027 research awardees1. Ensuring Intellectual Honesty Advancing LLM Reasoning Faithfulness without Faithfulness Rewards Faculty: Hao Peng Professor Peng is tackling the "hallucination" problem at its core. By developing methods to ensure large language models (LLMs) follow a logical, faithful reasoning path-without relying on traditional, often biased, reward systems-this work ensures that when an AI gives an answer, the "why" behind it is actually true, a critical component for trustworthy financial applications. TrACE-Contrast: Faithful & Consistent Code Reasoning via Trace-Aware Contrastive Learning Faculty: Talia Ringer and Reyhan Jabaarvand Writing code is one thing; understanding how it executes is another. Professor Ringer's project uses contrastive learning to align a model's code generation with its actual execution "trace." This ensures that AI-generated software is verified and logically sound, which is vital for maintaining the integrity of our core technology systems. 2. Bridging the Gap: Agency & Imagination Creative LLM Agents based on Thinking with Imagination Faculty: H

## Capital One Announces Open-Source VulnHunter Agentic AI Code Security Tool

DevFeed: [Capital One Announces Open-Source VulnHunter Agentic AI Code Security Tool](<https://devfeed.tech/articles/announcing-vulnhunter-22570.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/announcing-vulnhunter-ce9784834ca9?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-07-17T17:01:41Z

Content type: release

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [open-source-security](<https://devfeed.tech/topics/open-source-security.md>), [code security](<https://devfeed.tech/topics/code-security.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>), [Vulnerabilities](<https://devfeed.tech/topics/vulnerabilities.md>), [Developer Tools](<https://devfeed.tech/topics/developer-tools.md>), [Developer experience](<https://devfeed.tech/topics/developer-experience.md>)

Tags: [agentic-ai-security](<https://devfeed.tech/tags/agentic-ai-security.md>), [ai-code-security](<https://devfeed.tech/tags/ai-code-security.md>), [ai-security](<https://devfeed.tech/tags/ai-security.md>), [build](<https://devfeed.tech/tags/build.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [code-vulnerability](<https://devfeed.tech/tags/code-vulnerability.md>), [developer](<https://devfeed.tech/tags/developer.md>), [developer-experience](<https://devfeed.tech/tags/developer-experience.md>), [developers](<https://devfeed.tech/tags/developers.md>), [exploit](<https://devfeed.tech/tags/exploit.md>), [generative-ai-tools](<https://devfeed.tech/tags/generative-ai-tools.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [vulnerabilities](<https://devfeed.tech/tags/vulnerabilities.md>)

### AI overview

Capital One announces the open-source release of VulnHunter, an agentic AI security tool that analyzes source code from an attacker's perspective. It is designed to identify potentially exploitable defects, map prospective attack paths, and propose targeted code remediations.

### Source excerpt

Capital One's open-source, agentic AI code security tool. The rules of software security are changing faster than most defenders can keep pace. Advanced AI models have dramatically lowered the barrier for bad actors to discover and exploit vulnerabilities in software. What once required significant skill and time can now be automated, accelerated, and scaled. The world faces an increasingly short window of time before highly sophisticated, next-generation AI attack capabilities become affordable and accessible to virtually every adversary. Across the industry, organizations are racing to prepare for this paradigm shift. Traditional environmental protections like network segmentation, identity controls, and monitoring remain essential, but are no longer sufficient on their own. The ultimate defense in this new reality requires a shift in approach: organizations need to consider and detect the vulnerabilities in their code and fix them before adversaries can deploy advanced models to discover and exploit them. At Capital One, we decided that the right response to AI-enabled threats wasn't to wait, but to build cutting-edge AI-driven defenses and put them in the hands of defenders everywhere. That's why we are announcing today the open-source release of VulnHunter, an advanced agentic AI security tool designed to apply proactive, attacker-perspective analysis directly to the source code. Developed internally at Capital One, VulnHunter is not a traditional, passive vulnerability scanner. It represents a shift in defensive tooling with an agentic reasoning workflow to identify potentially exploitable defects, map prospective attack paths, and propose highly targeted code remediations. Built for the developer experience To fully unlock the utility of VulnHunter, we knew ease of use mattered. A persistent challenge with traditional security tools is that they are often built primarily to enforce rigid cybersecurity practices, without much consideration for a developer's ac

## Optimizing OPA performance: From arrays to objects

DevFeed: [Optimizing OPA performance: From arrays to objects](<https://devfeed.tech/articles/optimizing-opa-performance-from-arrays-to-objects-22577.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/optimizing-opa-performance-from-arrays-to-objects-a3c966acdaa5?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-07-07T14:25:30Z

Content type: article

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [Open Policy Agent](<https://devfeed.tech/topics/open-policy-agent.md>), [performance-engineering](<https://devfeed.tech/topics/performance-engineering.md>), [Algorithms](<https://devfeed.tech/topics/algorithms.md>), [computational complexity](<https://devfeed.tech/topics/computational-complexity.md>), [Data Management](<https://devfeed.tech/topics/data-management.md>), [data-engineering](<https://devfeed.tech/topics/data-engineering.md>), [anchore](<https://devfeed.tech/topics/anchore.md>)

Tags: [algorithms](<https://devfeed.tech/tags/algorithms.md>), [arrays](<https://devfeed.tech/tags/arrays.md>), [cloud-native](<https://devfeed.tech/tags/cloud-native.md>), [data](<https://devfeed.tech/tags/data.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [faster](<https://devfeed.tech/tags/faster.md>), [kubernetes](<https://devfeed.tech/tags/kubernetes.md>), [large](<https://devfeed.tech/tags/large.md>), [nested](<https://devfeed.tech/tags/nested.md>), [opa](<https://devfeed.tech/tags/opa.md>), [open-policy-agent](<https://devfeed.tech/tags/open-policy-agent.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [optimizing](<https://devfeed.tech/tags/optimizing.md>), [performance](<https://devfeed.tech/tags/performance.md>), [performance-engineering](<https://devfeed.tech/tags/performance-engineering.md>), [performance-tuning](<https://devfeed.tech/tags/performance-tuning.md>), [policies](<https://devfeed.tech/tags/policies.md>), [rego](<https://devfeed.tech/tags/rego.md>), [software-development](<https://devfeed.tech/tags/software-development.md>), [software-engineering](<https://devfeed.tech/tags/software-engineering.md>), [solution](<https://devfeed.tech/tags/solution.md>)

### AI overview

In an SBOM risk-enrichment example, replacing a 1.9-million-row EPSS array with a CVE-keyed object let Rego use direct object access instead of repeatedly traversing and comparing array entries. The JSON shrank from 46 MB to 20 MB. Processing 124 vulnerability profiles fell from about 48 seconds to 0.654 seconds (about 98% faster); query evaluation dropped from about 26 seconds to 5 milliseconds (about 99%). The article uses OPA's CLI profiler and metrics to locate the bottleneck, then transforms the data with jq and updates the policy to look up EPSS values by CVE ID. It notes that bundle loading and data parsing improved more modestly, and that further policy optimizations may be possible.

### Source excerpt

Achieve 99% faster Rego policy execution through optimization. Note: This post focuses on one aspect of performance tuning Rego policies and datasets evaluated by OPA-arrays vs. objects. The Rego Style Guide and Regal Rego linter are very helpful resources for learning Rego best practices and avoiding code smells in Rego policies. There is also the OPA performance tuning documentation. In 2018, I started using open policy agent (OPA) as a solution for controlling and preventing unwanted behaviors in our Kubernetes Clusters. OPA, along with Kubernetes Dynamic Admission Control, provided a means to build preventive controls. Since then, I have worked with several PaC solutions. I have always stayed close to the OPA tool set because of how well it supports multiple use cases. OPA is domain agnostic and can be used with virtually any use case, as long as you supply the correct data and policies. To that end, OPA use cases have expanded throughout several technical disciplines, such as cloud-native computing and software supply chain management. OPA performance engineering OPA enables us to unify PaC solutions across multiple use cases and systems, using the same languages and tools. However, there is always room for improvement and performance engineering policies and the execution thereof. In addition, optimizing data that policies evaluate and mutate should be part of our focus when we deliver OPA-based solutions. Recently I was asked to help with OPA performance issues. I made several recommendations, but I overlooked one simple and glaring issue: the poor performing policy was processing a large data set using nested-arrays, instead of the best practice of using keyed-objects. Later, something was bothering me about my interaction and I realized that while I gave decent architectural level advice, I completely missed the best engineering advice. Rego policies and data should be optimized just like other algorithms and relative data, and part of that optimization is

## LLM reasoning and agentic safety at ICML 2026

DevFeed: [LLM reasoning and agentic safety at ICML 2026](<https://devfeed.tech/articles/llm-reasoning-and-agentic-safety-at-icml-2026-22575.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/llm-reasoning-and-agentic-safety-at-icml-2026-55f341e21caa?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-07-02T14:14:40Z

Content type: article

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [LLM security](<https://devfeed.tech/topics/llm-security.md>), [ai safety](<https://devfeed.tech/topics/ai-safety.md>), [post-training](<https://devfeed.tech/topics/post-training.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [grpo](<https://devfeed.tech/topics/grpo.md>), [Post-training optimization](<https://devfeed.tech/topics/post-training-optimization.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [academic](<https://devfeed.tech/tags/academic.md>), [agent-safety](<https://devfeed.tech/tags/agent-safety.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-research](<https://devfeed.tech/tags/ai-research.md>), [conference](<https://devfeed.tech/tags/conference.md>), [deep-learning](<https://devfeed.tech/tags/deep-learning.md>), [distillation](<https://devfeed.tech/tags/distillation.md>), [execution](<https://devfeed.tech/tags/execution.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [framework](<https://devfeed.tech/tags/framework.md>), [frameworks](<https://devfeed.tech/tags/frameworks.md>), [general](<https://devfeed.tech/tags/general.md>), [generate](<https://devfeed.tech/tags/generate.md>), [icml](<https://devfeed.tech/tags/icml.md>), [icml-2026](<https://devfeed.tech/tags/icml-2026.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-reasoning](<https://devfeed.tech/tags/llm-reasoning.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [safety](<https://devfeed.tech/tags/safety.md>), [science](<https://devfeed.tech/tags/science.md>)

### AI overview

The excerpt presents Capital One research on reasoning, data generation, time-series evaluation, and agent safety at ICML 2026. Critique-Guided Distillation (CGD) trains models to refine flawed answers using teacher critiques that are absent at inference; across five model families it reports 7% average gains on mathematical reasoning and preserves instruction-following better than Critique Fine-Tuning. EPSVec sanitizes dataset steering vectors once, decoupling its privacy budget from synthetic text generation. BEDTime evaluates time-series description tasks and finds vision-language models capable, dedicated time-series models underperforming, and language-only methods weakest among 13 models tested. TRACER combines surprisal, repetition, situational awareness, and tool coherence signals into trajectory-level risk estimates; it reports improved failure detection and selective execution, often flagging failures in the first 20% of a trajectory. Workshop work includes interpretable binary-question evaluation, active tool-assisted fault diagnosis, and entropy-guided distillation. These are research-reported results and claims, not independent assessments.

### Source excerpt

Explore our latest research in critique-guided distillation and multi-turn agent uncertainty in Seoul.Explore our latest research in critique-guided distillation and multi-turn agent uncertainty in Seoul. Capital One technologists are excited to participate in the 43rd International Conference on Machine Learning (ICML) taking place at the COEX Convention & Exhibition Center in Seoul, South Korea, July 6-11, 2026. As a premier global venue for machine learning research, ICML provides an essential forum for exploring foundational advancements, algorithmic innovations and cutting-edge deep learning systems. Capital One is excited to share advancements in large language model (LLM) scaling efficiencies, multi-turn tool-using agent safety and the development of robust, trustworthy AI frameworks. This work delivers the underlying engineering and algorithmic improvements crucial for deploying the next generation of safe financial technologies. Main conference research: Robust reasoning and agentic risk The following research, accepted to the ICML Main Conference, pushes the boundaries of how models self-correct, how trajectory-level risks can be proactively flagged, and how multi-turn agent interactions maintain reliable execution. This section features work led by Capital One researchers alongside deep collaborations with academic partners. Critique-Guided Distillation for Robust Reasoning via Refinement Capital One Authors: Berkcan Kapusuzoglu, Supriyo Chakraborty, Michael Lee, Sambit Sahu Supervised fine-tuning with expert demonstrations often produces models that imitate outputs without internalizing the reasoning processes needed for robust generalization. While critique-based approaches show promise, training models to generate critiques directly, such as Critique Fine-Tuning (CFT), can lead to output-format drift and degradation of general capabilities. We propose Critique-Guided Distillation (CGD), a training framework that decouples critique consumption from crit

## Capital One at ACL 2026

DevFeed: [Capital One at ACL 2026](<https://devfeed.tech/articles/capital-one-at-acl-2026-22571.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/capital-one-at-acl-2026-ad9c245333fe?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-07-01T15:28:51Z

Content type: news

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [LLM security](<https://devfeed.tech/topics/llm-security.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Security Attacks](<https://devfeed.tech/topics/security-attacks.md>), [Security research](<https://devfeed.tech/topics/security-research.md>), [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [dataset](<https://devfeed.tech/topics/dataset.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [academic](<https://devfeed.tech/tags/academic.md>), [academic-partnerships](<https://devfeed.tech/tags/academic-partnerships.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-ml](<https://devfeed.tech/tags/ai-ml.md>), [ai-research](<https://devfeed.tech/tags/ai-research.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-red-teaming](<https://devfeed.tech/tags/llm-red-teaming.md>), [natural-language-processing](<https://devfeed.tech/tags/natural-language-processing.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [routing](<https://devfeed.tech/tags/routing.md>), [science](<https://devfeed.tech/tags/science.md>)

### AI overview

The excerpt describes two Capital One ACL 2026 papers. Adaptive Instruction Composition applies reinforcement learning and a neural contextual bandit to vary red-team attack strategies, aiming to balance diversity and effectiveness in finding LLM vulnerabilities. Routing with Generated Data (RGD), a collaboration with the University of North Carolina, Chapel Hill, proposes CASCAL, a query-only router that estimates model correctness through consensus voting and hierarchical clustering; the excerpt reports it outperforms traditional baselines even with weaker generator datasets. The excerpt cuts off during a third paper's description, so it does not support an overview of the full research program.

### Source excerpt

Discover how Capital One is advancing state-of-the-art AI/ML science through collaborative natural language processing research.Advancing AI and NLP Frontiers at ACL 2026 As language models grow more deeply integrated into technology ecosystems, pioneering robust, efficient, and reliable Natural Language Processing (NLP) techniques becomes paramount. Capital One continues to invest in state-of-the-art AI/ML science through deep multi-sector collaboration and peer-reviewed research. At the upcoming Annual Meeting of the Association for Computational Linguistics (ACL 2026), Capital One researchers and academic partners will showcase novel findings stretching from LLM security to multilingual capabilities. Through the Science & Academic Partnerships program, Capital One bridges industry needs with academic expertise, funding critical university research and engineering solutions that make technology safer and more powerful. Our accepted publications at ACL 2026 demonstrate this thriving flywheel of talent and collaborative innovation across multiple research categories. Main Conference Research Adaptive Instruction Composition for Automated LLM Red Teaming Routing with Generated Data: Annotation-Free LLM Skill Estimation and Expert Selection Capital One Authors: Jesse Zymet, Swapnil Shinde, Sahil Wadhwa, Andy Luo Overview: Standard red teaming approaches often struggle with a limited range of jailbreak strategies or rely on ineffective, randomized crowd-sourced tactics. This paper introduces a novel framework -- Adaptive Instruction Composition -- that utilizes reinforcement learning and a neural contextual bandit to tailor attack compositions dynamically, balancing diversity and effectiveness to proactively uncover target model vulnerabilities. Routing with Generated Data: Annotation-Free LLM Skill Estimation and Expert Selection Capital One Authors: Genta Winata, Sambit Sahu, Supriyo Chakraborty, Shixiong Zhang Overview: Emerging from our gifted research collaboration

## New Open-Source Context Specs Released

DevFeed: [New Open-Source Context Specs Released](<https://devfeed.tech/articles/new-open-source-context-specs-released-22576.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/new-open-source-context-specs-released-1f65bfa9db9f?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-06-24T16:58:13Z

Content type: article

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [Spec Driven Development](<https://devfeed.tech/topics/spec-driven-development.md>), [AI-assisted coding](<https://devfeed.tech/topics/ai-assisted-coding.md>), [context window](<https://devfeed.tech/topics/context-window.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>), [software-development](<https://devfeed.tech/topics/software-development.md>)

Tags: [ai-assisted-coding](<https://devfeed.tech/tags/ai-assisted-coding.md>), [context-engineering](<https://devfeed.tech/tags/context-engineering.md>), [context-window](<https://devfeed.tech/tags/context-window.md>), [development](<https://devfeed.tech/tags/development.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [software-engineering](<https://devfeed.tech/tags/software-engineering.md>), [spec-driven-development](<https://devfeed.tech/tags/spec-driven-development.md>)

### AI overview

Capital One has open-sourced Context Specs, a spec-driven development framework that uses context engineering for AI-assisted coding. It captures domain knowledge in reusable experts and uses automated feedback signals to verify an agent's work, addressing context-window constraints such as context decay, pollution, and compaction loss.

### Source excerpt

Capital One open-sources Context Specs, an SDD framework using context engineering to bridge business ideas and code. Every developer has had the same experience: The AI writes code confidently, only for you to realize that it has missed the mark. The fix isn't a better model, it's better context. Right context, right time. That's the whole game. To address this challenge, Capital One is open-sourcing Context Specs, a framework for spec-driven development (SDD) that treats context engineering-the practice of building systems that dynamically decide what your agent sees and when it sees it-as the primary lever for AI-assisted coding. The framework achieves this by capturing a team's domain-specific knowledge into reusable "experts" that you create once and compose across the entire workflow, and then using automated feedback "signals" to verify the agent's work. The real problem: Your agent's context window is finite Every AI coding agent operates within a context window, a fixed amount of information it can see at any given moment. This constraint creates three mechanical failure modes that plague every developer using AI tools today: Context decay: Older messages in a conversation get ignored, summarized or dropped entirely. That careful instruction you gave 20 messages ago? Gone. Context pollution: When an agent searches your codebase autonomously, it pulls in irrelevant files. Every irrelevant token displaces a useful one. Compaction loss: When the window fills up, the system summarizes history to make room. You don't control what gets dropped. These aren't edge cases; they're the default experience. Most frameworks for AI-assisted development weren't designed with these constraints in mind. Some frameworks generate thousands of lines of specification markdown before a single line of code is written, burning millions of tokens on ceremony. Other frameworks are so lightweight they leave everything to interpretation, forcing the agent to search blindly and pollute

## DataAgents: How we turned 9 months of analysis into 10 days

DevFeed: [DataAgents: How we turned 9 months of analysis into 10 days](<https://devfeed.tech/articles/dataagents-how-we-turned-9-months-of-analysis-into-10-days-22572.md>)

Original publisher: [Read original article](<https://medium.com/capital-one-tech/dataagents-how-we-turned-9-months-of-analysis-into-10-days-8d6ed482f5d7?source=rss----3db3a67cb648---4>)

Author: Capital One Tech

Published: 2026-06-09T22:42:18Z

Content type: article

Language: en

Sources: [Capital One Tech](<https://devfeed.tech/sources/capital-one-tech.md>)

Topics: [Data Quality](<https://devfeed.tech/topics/data-quality.md>), [knowledge-engineering](<https://devfeed.tech/topics/knowledge-engineering.md>), [Documentation](<https://devfeed.tech/topics/documentation.md>), [Tool](<https://devfeed.tech/topics/tool.md>)

Tags: [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [ai](<https://devfeed.tech/tags/ai.md>), [amazon](<https://devfeed.tech/tags/amazon.md>), [amazon-s3](<https://devfeed.tech/tags/amazon-s3.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [aws](<https://devfeed.tech/tags/aws.md>), [azure](<https://devfeed.tech/tags/azure.md>), [behavior](<https://devfeed.tech/tags/behavior.md>), [catalog](<https://devfeed.tech/tags/catalog.md>), [classification](<https://devfeed.tech/tags/classification.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [compliance](<https://devfeed.tech/tags/compliance.md>), [confidence](<https://devfeed.tech/tags/confidence.md>), [data](<https://devfeed.tech/tags/data.md>), [data-analysis](<https://devfeed.tech/tags/data-analysis.md>), [data-governance](<https://devfeed.tech/tags/data-governance.md>), [days](<https://devfeed.tech/tags/days.md>), [deep-dive](<https://devfeed.tech/tags/deep-dive.md>), [detection](<https://devfeed.tech/tags/detection.md>), [ec2](<https://devfeed.tech/tags/ec2.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [expert-analysis](<https://devfeed.tech/tags/expert-analysis.md>), [false-positives](<https://devfeed.tech/tags/false-positives.md>)

### AI overview

DataAgents cut cloud resource dormancy analysis from an estimated 6-9 months to 10 days by pairing an AI agent with a standardized, authoritative inventory covering about 350 resource types across AWS, Azure, and GCP. The process first sorts types by whether configuration data can reveal dormancy or telemetry is needed, then generates Spark SQL detection logic with confidence ratings, false-positive risks, and explanations. Human reviewers individually recheck medium- and low-confidence cases; this validation caught incorrect field names and overconfident classifications before production. The analysis found that explicit states such as stopped or unattached support higher-confidence detection, while age-only rules can produce 40-60% false positives for resources used without configuration changes. The authors report that 12 high-confidence types account for 30-40% of potential dormancy savings. They present a trustworthy, well-documented data source and domain-aware human review as prerequisites; the pattern may extend to other large catalogs with entity-specific rules.

### Source excerpt

An engineering deep dive into the pattern that changed how we approach large-scale classification problems. Every engineering team has that project sitting in the backlog. The one where someone says, "We really should analyze all of these," and the room goes quiet. Everyone knows what "all of these" means -- hundreds of entities, complex rules, no clear starting point. For us, it was cloud resource dormancy detection. We had around 350 distinct cloud resource types spread across AWS, Azure and Google Cloud Platform (GCP). Each type has different behavior patterns. An EC2 instance sitting idle looks nothing like a dormant Amazon S3 (S3) bucket or an unattached Elastic IP. Detecting dormancy required understanding what "active" means for each specific resource, then writing detection logic that wouldn't flood operations teams with false positives. Traditional estimate: 6-9 months of expert analysis. Actual time: 10 days. Here's how we did it, and more importantly, here's the reusable pattern behind it. The problem with large-scale analysis: Before we get to the solution, it's worth naming the pattern that makes these projects so painful. It shows up everywhere: Cloud resources - Which of our 350 resource types are dormant? Data governance - Which of our 800 tables have quality issues we should monitor? Security - Which of our access entitlements violate least-privilege principles? Compliance - Which of our 500 policy controls need remediation? In every case, the structure is similar -- a large catalog of heterogeneous entities, entity-specific rules that don't generalize, unknown priorities and a high cost for getting it wrong. The traditional approach is not just slow. It's structurally limited. You get coverage of the "obvious" cases, inconsistent logic across analysts, and tribal knowledge that evaporates when people leave. What you need is something that can assess each entity, apply consistent criteria, prioritize by confidence and document its reasoning. The DataA