# retrieval Published articles for retrieval. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## Glyph: A Multi-Strategy Agentic System for Column Description and Sensitivity-Ontology Tagging of Enterprise Data Catalogs DevFeed: [Glyph: A Multi-Strategy Agentic System for Column Description and Sensitivity-Ontology Tagging of Enterprise Data Catalogs]() Original publisher: [Read original article]() Published: 2026-09-16T00:00:00Z Content type: article Language: en Sources: [Apple Machine Learning Research]() Topics: [data](), [Large Language Model](), [Code](), [Retrieval-Augmented Generation](), [Access Control](), [Graphs](), [GitHub]() Tags: [agents](), [classification](), [code](), [data](), [documentation](), [governance](), [production](), [provenance](), [retrieval](), [vector-database]() ### AI overview Glyph is a production system for generating column descriptions and assigning sensitivity-ontology labels in enterprise data catalogs. It uses cooperating LLM agents, source-code-grounded retrieval, parallel tagging strategies, vector-based metadata matching, and ranked-output fusion to support auditable cataloging. ### Source excerpt Enterprise data lakes accumulate tables faster than human stewards can document or classify them, leaving columns with missing descriptions and unassigned governance labels. This documentation debt undermines data discovery, access control, and regulatory compliance. We present Glyph, a production system that frames two coupled problems, column description generation and column type annotation for data classification, as cooperating LLM agents orchestrated as stateful graphs. The Descriptor grounds generation in the pipeline source code that produces each column, retrieved on demand from an... ## Bypassing inference bottlenecks: Accelerating complex AI search with Retrieve-for-Train DevFeed: [Bypassing inference bottlenecks: Accelerating complex AI search with Retrieve-for-Train]() Original publisher: [Read original article]() Published: 2026-09-15T20:00:35Z Content type: article Language: en Sources: [The latest research from Google]() Topics: [AI search](), [Reinforcement learning](), [Artificial Intelligence](), [Google](), [LLMs](), [Algorithms & Theory]() Tags: [ai](), [ai-search](), [algorithms-theory](), [data-mining-modeling](), [diffusion](), [generative-ai](), [google](), [icml](), [icml-2026](), [reinforcement-learning](), [research](), [retrieval](), [rl](), [search]() ### AI overview Google Research presents Retrieve-for-Train, a framework that uses offline reinforcement learning to compile reward-aligned query fan-outs into training data for a lightweight diffusion retriever. The approach is intended to produce diverse, complementary, and coherent search-result sets in a single inference pass, reducing reliance on expensive inference-time reasoning. ### Source excerpt Algorithms & Theory ## Seagate and WD AI Storage Research Finds Enterprises Rank Storage Above Compute as the AI Bottleneck DevFeed: [Seagate and WD AI Storage Research Finds Enterprises Rank Storage Above Compute as the AI Bottleneck]() Original publisher: [Read original article]() Author: Lyle Smith Published: 2026-09-15T17:23:54Z Content type: news Language: en Sources: [StorageReview.com]() Topics: [Data Infrastructure](), [AI Infrastructure](), [idc](), [Inference](), [Retrieval-Augmented Generation](), [synthetic-data]() Tags: [ai](), [compute](), [data-infrastructure](), [datasets](), [enterprise](), [genai](), [hdd](), [idc](), [inference](), [reports](), [research](), [retrieval](), [storage](), [synthetic-data]() ### AI overview Seagate and WD published separate studies indicating that AI is increasing enterprise storage requirements and extending data retention. Although their headline percentages differ because they asked different questions, both reports point to storage becoming a larger part of AI infrastructure planning alongside growing archive and retrieval needs. ### Source excerpt Seagate and WD published separate AI storage studies within days of each other; the headline numbers: Seagate says 99% of enterprises expect AI to increase their storage requirements over the next three years, while WD's IDC research puts the comparable figure at 74%. Read the fine print, and both reports land in the same directional The post Seagate and WD AI Storage Research Finds Enterprises Rank Storage Above Compute as the AI Bottleneck appeared first on StorageReview.com. ## Unified Knowledge Graph RAG on AWS: GraphRAG and LightRAG on one stack DevFeed: [Unified Knowledge Graph RAG on AWS: GraphRAG and LightRAG on one stack]() Original publisher: [Read original article]() Author: Jonas Kim Published: 2026-09-14T16:55:58Z Content type: article Language: en Sources: [AWS Open Source Blog]() Topics: [Retrieval Augmented Generation (RAG)](), [Graphs](), [Amazon Bedrock](), [Amazon OpenSearch Service](), [Amazon Web Services]() Tags: [amazon-bedrock](), [amazon-opensearch-service](), [aws](), [knowledge-graph](), [open-source](), [rag](), [retrieval](), [search](), [vector-search]() ### AI overview This article presents a unified knowledge-graph RAG stack that brings Microsoft GraphRAG and HKUDS LightRAG together on Amazon Bedrock, Amazon Neptune, and Amazon OpenSearch Service. The shared stack supports common ingestion, indexing, caching, and multilingual handling while allowing the retrieval methodology to be selected per query. ### Source excerpt Picture a compliance analyst staring at a few thousand contracts, amendments, and internal memos, trying to answer one question that sounds straightforward: "Which of our obligations are exposed if this one milestone slips?" The answer isn't written in any single document. It's stitched across three -- a master agreement that ties a payment to a [...] ## OpenSearch Wins Analytics & Data Intelligence Solutions Category in the SiliconANGLE TechForward Awards DevFeed: [OpenSearch Wins Analytics & Data Intelligence Solutions Category in the SiliconANGLE TechForward Awards]() Original publisher: [Read original article]() Author: Kristi Piechnik Published: 2026-09-14T12:00:14Z Content type: news Language: en Sources: [OpenSearch]() Topics: [Open Source](), [AI Infrastructure](), [observability](), [AI search](), [Security](), [Retrieval Augmented Generation (RAG)](), [Databases](), [real-time]() Tags: [2026](), [agentic-ai](), [ai-agents](), [ai-infrastructure](), [announcements](), [awards](), [data](), [enterprise](), [observability](), [open-source](), [opensearch](), [real-time](), [recognition](), [retrieval](), [retrieval-augmented-generation-rag](), [search](), [security]() ### AI overview OpenSearch won the Analytics & Data Intelligence Solutions category in SiliconANGLE Media's 2026 TechForward Awards. The recognition highlights its open source, vendor-neutral platform for enterprise search, observability, security analytics, vector databases, and agentic AI workloads. ### Source excerpt Recognition validates open source momentum, architectural consolidation, and enterprise scale as the project marks five years of community growth The post OpenSearch Wins Analytics & Data Intelligence Solutions Category in the SiliconANGLE TechForward Awards appeared first on OpenSearch. ## Enterprise-Grade Precision for Long-Context Multimodal Embedding Inference on Cloud TPU DevFeed: [Enterprise-Grade Precision for Long-Context Multimodal Embedding Inference on Cloud TPU]() Original publisher: [Read original article]() Author: Anthony Su; Injae Kwak Published: 2026-09-12T11:04:33.891311Z Content type: article Language: en Sources: [Google Developers Blog]() Topics: [Embeddings](), [vllm](), [Inference](), [long-context](), [autoscaling](), [Kubernetes](), [Cloud](), [Google](), [Artificial Intelligence](), [model-deployment](), [Open Source]() Tags: [autoscaling](), [embedding](), [github](), [gpu](), [inference](), [kubernetes](), [long-context](), [multimodal](), [open-source](), [retrieval](), [tpu](), [vllm]() ### AI overview This article explains how Google Cloud integrated native TPU support into vLLM to serve long-context, multimodal embedding models at production scale. It describes elastic TPU provisioning with Google Kubernetes Engine, autoscaling across accelerator types, and TPU-specific optimizations for long sequences and chunked prefill. The resulting setup is designed to preserve numerical parity with GPU reference baselines while supporting high-throughput semantic retrieval applications. ### Source excerpt Google Cloud has natively integrated TPU support into the vLLM serving engine, allowing developers to elastically scale high-demand embedding pipelines using Google Kubernetes Engine (GKE). To handle massive 15K+ token contexts for models like Qwen3-Embedding-8B, the engineering team implemented TPU-specific optimizations such as hardware-safe tensor alignment, JAX/XLA compilation pre-warming, and a hybrid StepPool architecture for chunked prefill management. These enhancements achieve near-perfect numerical parity with reference GPU baselines, and developers can immediately leverage the open-sourced setup recipes on the AI-Hypercomputer GitHub to build their own high-throughput semantic retrieval applications. ## Evolving Pinterest's Embedding Retrieval Platform DevFeed: [Evolving Pinterest's Embedding Retrieval Platform]() Original publisher: [Read original article]() Author: Pinterest Engineering Published: 2026-09-11T15:01:03Z Content type: article Language: en Sources: [Pinterest Engineering Blog - Medium]() Topics: [Architecture & Design](), [Algorithms](), [Scalability](), [Latency](), [IO]() Tags: [ann](), [cost](), [embedding](), [embeddings](), [engineering](), [infrastructure](), [latency](), [memory](), [models](), [pinterest](), [platform](), [quantization](), [real-time](), [retrieval](), [scalability](), [search]() ### AI overview Pinterest describes evolving its Manas embedding-retrieval platform to address the cost, scale, and flexibility challenges of serving billions of embeddings. The excerpt covers ANN search, vector quantization, and SSD-based serving. ### Source excerpt Authors: Bowen Zhou | Staff Software Engineer; Shan Gao | Senior Software Engineer; Jingwen Hu | Software Engineer II; Wenjiang Chu | Staff Software Engineer The Billion-Embedding Challenge At Pinterest, the "signal" is our lifeblood. Whether it's a home decor enthusiast finding the perfect rug or a fashion seeker discovering a new aesthetic, our discovery engine relies on understanding deep semantic relationships to help our users find inspirations. Over the last few years, the explosive growth of embedding-based retrieval has fundamentally transformed how we surface these signals -- and at the heart of that transformation is Manas, Pinterest's in-house distributed search platform. Embedding Retrieval is one of the core capabilities of Manas, supporting multiple approximate nearest neighbor search algorithms, hybrid queries with both token and embedding clauses, as well as real-time updates to ensure fresh contents become searchable within seconds. Deployed on over 80 clusters and serving billions of embeddings, Manas embedding retrieval powers all major product surfaces at Pinterest including Home Feed, Search, Related Pins, Ads, and Notifications. However, as our corpus scales toward tens of billions of embeddings and our models capture increasingly complex interactions, we face mounting challenges around cost efficiency, scalability, and flexibility. On the infrastructure side, traditional ANN algorithms like HNSW are notoriously memory-hungry -- they require the entire index to reside in RAM to maintain low query latency, making cost grow linearly with corpus size. On the modeling side, the classic two-tower retrieval paradigm is too restrictive: it reduces each candidate to a single embedding and scores relevance through a simple dot product, leaving little room to express richer, context-dependent notions of similarity. To tackle these challenges, our team has been evolving Manas's embedding retrieval stack across three fronts: Quantization. We reduce the memor ## Three principles for building a vector platform at Thumbtack DevFeed: [Three principles for building a vector platform at Thumbtack]() Original publisher: [Read original article]() Author: John Zhu Published: 2026-09-10T15:45:00Z Content type: article Language: en Sources: [Thumbtack Engineering - Medium]() Topics: [Embeddings](), [AI search](), [Database](), [data](), [Machine learning](), [API]() Tags: [api](), [data](), [embeddings](), [engineering](), [etl](), [infrastructure](), [machine-learning](), [ml-platform](), [pgvector](), [retrieval](), [search](), [vector](), [vector-search]() ### AI overview This article explains how Thumbtack built a vector platform that lets ML engineers deploy production vector search without managing database access, custom ETL, or query services. It describes three guiding principles: reuse existing infrastructure, treat embeddings as data, and reduce adoption costs for future teams. ### Source excerpt Reusing what we already had, treating embeddings as data, and lowering the next team's cost Today, an ML engineer at Thumbtack can stand up production vector search without negotiating database access, building a custom ETL, or writing a query service. The team brings their choice of embedding model, the data, and the query; the platform handles what connects them. It took several iterations to get to this point. In this post we'll walk through how we got there and the three principles that shaped what we built. A vector database stores high-dimensional numeric arrays (embeddings) and serves nearest-neighbor queries against them. It's how an ML system asks "what's most similar to this?" instead of "what matches this exact key?" The shift from exact lookup to semantic retrieval is what makes vectors useful: a search can return results that mean the same thing, not just results that spell the same. At Thumbtack, embeddings sit between the models that produce them and the services that consume them: language models for text, multimodal models for images, retrieval models for ranking. The platform we describe here is where those embeddings live and how teams reach for them when they need to. Three principles shaped what we built. Reuse what we have: extend the infrastructure we already run rather than stand up a new system. Treat embeddings as data: flow them through the same pipelines that move every other dataset at the company. Lower the next team's cost: make the platform easier to adopt than to work around. Each principle shaped one layer of the system, and together they took vector search from a one-off project to a platform that any team can build on. Architecture at a glance The platform has four moving parts: where embeddings come from, how they reach the database, where they live, and how consumers query them. Each is a layer, and together they form a pipeline that produces vectors and serves similarity searches as a typed API call. The diagram below traces a ## Introducing ChatGPT for Financial Services DevFeed: [Introducing ChatGPT for Financial Services]() Original publisher: [Read original article]() Published: 2026-09-10T07:00:00Z Content type: article Language: en Sources: [OpenAI News]() Topics: [data](), [datasets](), [Model Context Protocol](), [Latency]() Tags: [chatgpt](), [data](), [datasets](), [financial-services](), [gpt](), [latency](), [mcp](), [models](), [openai](), [product](), [reasoning](), [retrieval](), [security]() ### AI overview ChatGPT for Financial Services combines GPT-6 Astra with built-in, OpenAI-hosted financial data for research, financial modeling, and client materials. It includes data-provider datasets, citations, enterprise controls, and MCP performance improvements. ### Source excerpt Introducing ChatGPT for Financial Services, combining built-in financial data and GPT-6 Astra for research, modeling, and client-ready materials. ## Build a support chatbot with Appwrite Functions and VectorsDB DevFeed: [Build a support chatbot with Appwrite Functions and VectorsDB]() Original publisher: [Read original article]() Author: Atharva Deosthale Published: 2026-09-10T00:00:00Z Content type: tutorial Language: en Sources: [Appwrite Blog]() Topics: [Appwrite](), [Chat Bot](), [Retrieval-Augmented Generation](), [Embeddings](), [React](), [SDKs](), [Node.js]() Tags: [api](), [build](), [chat](), [deploy](), [embedding](), [functions](), [node-js](), [rag](), [react](), [retrieval](), [support](), [tutorials]() ### AI overview This tutorial builds Harbor Help, a support chatbot for a fictional project-management product. It uses Appwrite VectorsDB to store help-center article embeddings, an Appwrite Function to retrieve relevant excerpts and generate answers through OpenRouter, and Appwrite Sites to host a React interface. The chatbot uses retrieval-augmented generation and can refuse questions outside the help center. ### Source excerpt Seed a help center, retrieve relevant articles with VectorsDB, and answer customer questions through an Appwrite Function. Deploy the chat interface on Appwrite Sites. ## Adaptive Instructed-Retriever: Frontier-Quality Search at 2x Lower Latency DevFeed: [Adaptive Instructed-Retriever: Frontier-Quality Search at 2x Lower Latency]() Original publisher: [Read original article]() Author: Cindy Wang; Cheng Li; Jialu Liu; Sean Kulinski; Arnav Singhvi; Wen Sun; Michael Bendersky Published: 2026-09-09T13:30:00Z Content type: article Language: en Sources: [Databricks]() Topics: [databricks](), [Latency](), [Low Latency](), [data]() Tags: [agents](), [announcements](), [benchmarks](), [cost](), [data](), [databricks](), [enterprise](), [latency](), [low-latency](), [performance](), [platform](), [retrieval](), [search](), [speed](), [third-party]() ### AI overview Databricks introduces Adaptive Instructed-Retriever, a retrieval model that combines fast parallel search with sequential multi-step search for harder enterprise queries. It adaptively spends extra computation only when useful, achieving comparable quality to leading third-party models at twice lower latency while improving over single-step retrieval on reported benchmarks. ### Source excerpt Effective enterprise data agents require search that is both accurate and fast. Earlier... ## Progressive Authentication for Production-Ready RAG Architectures DevFeed: [Progressive Authentication for Production-Ready RAG Architectures]() Original publisher: [Read original article]() Author: Aishwarya Murali Padikkal Published: 2026-09-08T07:16:00Z Content type: article Language: en Sources: [Data Science at Microsoft]() Topics: [Authentication](), [Retrieval Augmented Generation (RAG)](), [sensitive data](), [User Experience](), [Security]() Tags: [ai](), [ai-agent](), [authentication](), [data-science](), [privacy](), [rag](), [retrieval](), [sensitive-data](), [user-experience]() ### AI overview The article presents progressive authentication for RAG systems: answer public questions without sign-in, then authenticate within the conversation when a request requires access to sensitive personal data. It argues that static, upfront authentication does not fit conversations that shift between public and private contexts. ### Source excerpt Designing a production-ready RAG architecture where authentication becomes part of the conversation, not a prerequisite for it. By Aishwarya Murali Padikkal & Priyank Solanki From understanding to retrieval, grounding, and response, authentication enters only when the conversation requires access to sensitive data- illustrating the core idea of progressive authentication. Illustration generated using ChatGPT and refined by the authors. Two questions, one help line Imagine two taxpayers opening the same AI-powered tax assistant. The first asks, "When is the filing deadline this year?" The second asks, "What was my refund last year?" Both questions are valid. Both deserve fast, helpful answers. Yet they represent fundamentally different types of requests. Figure 1. The same assistant, two different journeys. Requests are routed according to the type of information being accessed. Illustration generated by AI (ChatGPT) and refined by the authors. The answer to the first lives in a public handbook. Anyone should be able to access it instantly- no sign-in, no friction. The answer to the second is within an individual's private tax records. Before revealing a single detail, the assistant must be certain it is interacting with the right person. Making public information instantly accessible while ensuring private information is revealed only to the right person, all within the same conversation, is the challenge we set out to solve. It is also where conventional RAG systems quietly fall apart. Traditional RAG architectures assume a much simpler interaction model: a user asks a question, the system retrieves relevant documents, and the model generates an answer. Authentication, if it exists at all, is treated as a one-time checkpoint at the start of the interaction. Real conversations are far less predictable. Users naturally shift contexts, starting with a public question, moving to a personal one, and then asking follow-up questions that rely on both. The conversation evo ## Vespa Newsletter, September 2026 DevFeed: [Vespa Newsletter, September 2026]() Original publisher: [Read original article]() Author: Bonnie Chase Published: 2026-09-08T00:00:00Z Content type: news Language: en Sources: [Vespa Blog]() Topics: [ann](), [Latency](), [Provisioning](), [Algorithm](), [telemetry](), [Graphs]() Tags: [2026](), [algorithm](), [cloud](), [features](), [graph](), [latency](), [newsletter](), [product](), [provisioning](), [ranking](), [retrieval](), [september-2026](), [telemetry](), [updates]() ### AI overview The September 2026 Vespa newsletter announces updates including time-constrained ANN search, sub-query ranking support, flexible provisioning, new rank features, and telemetry export. It also introduces Vespa.ai Live, an in-person community meetup focused on retrieval and ranking systems. ### Source excerpt Advances in Vespa include time-constrained ANN search, sub-query ranking support, flexible provisioning, new rank features and telemetry export ## Building a Memory-Driven Agent with NVIDIA NemoClaw DevFeed: [Building a Memory-Driven Agent with NVIDIA NemoClaw]() Original publisher: [Read original article]() Author: Tanya Lenz Published: 2026-09-04T18:04:55Z Content type: tutorial Language: en Sources: [NVIDIA Developer](), [NVIDIA Technical Blog]() Topics: [Architecture & Design](), [Authorization](), [Markdown]() Tags: [agentic-ai-generative-ai](), [ai-agent](), [build-ai-agents](), [developer-tools-techniques](), [enterprise](), [llms](), [memory](), [nemoclaw](), [openshell](), [retrieval](), [retrieval-augmented-generation-rag](), [security]() ### AI overview This tutorial describes building a memory-driven AI agent with NVIDIA NemoClaw for enterprise work. It presents a structured self model, separates evidence from derived knowledge and governed execution, and emphasizes retrieval, user corrections, security, and authorization. ### Source excerpt Enterprise work spans messages, decisions, projects, and obligations that change over time. An AI agent that starts without this context must reconstruct it... ## Ling 3.0 Flash Sante is now available on AI Gateway for free DevFeed: [Ling 3.0 Flash Sante is now available on AI Gateway for free]() Original publisher: [Read original article]() Author: Jerilyn Zheng Published: 2026-09-04T00:00:00Z Content type: release Language: en Sources: [Vercel News]() Topics: [Artificial Intelligence](), [context window]() Tags: [agentic](), [ai](), [ai-gateway](), [coding](), [context-window](), [free](), [healthcare](), [model](), [retrieval]() ### AI overview Ling 3.0 Flash Sante, a healthcare-focused model from inclusionAI, is available on AI Gateway for free through October 4. It supports medical reasoning, research, evidence retrieval, multi-step healthcare tasks, function calling, and a 256K-token context window. ### Source excerpt Ling 3.0 Flash Sante from inclusionAI is now available on AI Gateway, and free through October 4. Ling 3.0 Flash Sante is the healthcare-focused version of Ling 3.0 Flash. It's built for medical reasoning, research, evidence retrieval, and multi-step healthcare tasks. It has a 256K-token context window, supports function calling, and retains the base model's general reasoning, coding, and agentic capabilities. Choose a model ID based on what should happen after the free period: inclusionai/ling-3.0-flash-sante continues serving requests at standard rates after October 4. inclusionai/ling-3.0-flash-sante-free stops serving requests after October 4. Free requests remain visible in your usage and traces but show a cost of $0. Use the model: Try Ling 3.0 Flash Sante in the model playground, or see the free model page. Read more ## NeoMME: an efficient Multimodal-native and Multilingual Encoder DevFeed: [NeoMME: an efficient Multimodal-native and Multilingual Encoder]() Original publisher: [Read original article]() Author: Tony Wu; Aurélien Lac Published: 2026-09-03T13:13:48Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [vlm](), [Fine-tuning](), [Large Language Model](), [GPU]() Tags: [apache](), [architecture](), [diffusion](), [embeddings](), [fine-tuning](), [gpu](), [hugging-face](), [multimodal](), [quantization](), [retrieval](), [training](), [transformers](), [vector](), [vision](), [vlm]() ### AI overview NeoMME is a family of multilingual multimodal encoders trained from scratch with a masked discrete-diffusion objective. It uses one bidirectional Transformer for text tokens and image patches, and is fine-tuned for visual document retrieval with dense and late-interaction embeddings. ### Source excerpt We introduce NeoMME, a family of 260M and 800M multilingual multimodal encoders. Unlike many generative visual language models, NeoMME does not use a separate pretrained vision tower or a causal language model. A single bidirectional Transformer processes both text tokens and raw image patches, and we train the entire model from scratch with a masked discrete-diffusion objective. We fine-tuned NeoMME for visual document retrieval using ColPali's page-image approach. ## Give Your Coding Agents a Memory You Own DevFeed: [Give Your Coding Agents a Memory You Own]() Original publisher: [Read original article]() Author: David Corvoysier Published: 2026-09-03T00:00:00Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [AI-assisted coding](), [Traces](), [Inference](), [dataset]() Tags: [agents](), [announcement](), [coding](), [datasets](), [embedding](), [guide](), [inference](), [local](), [memory](), [open-source](), [retrieval](), [traces]() ### AI overview funes is a local, durable memory layer for coding agents that indexes prior session traces so agents can retrieve past decisions with provenance. It uses a deterministic pipeline with vector and BM25 search, reranking, recency weighting, and local storage. ### Source excerpt We're on a journey to advance and democratize artificial intelligence through open source and open science. ## Deploying a RAG Chatbot with Shared State and Storage Across Replicas DevFeed: [Deploying a RAG Chatbot with Shared State and Storage Across Replicas]() Original publisher: [Read original article]() Author: Avi Chawla Published: 2026-09-01T21:11:23Z Content type: tutorial Language: en Sources: [Daily Dose of Data Science]() Topics: [Retrieval-Augmented Generation](), [Kubernetes](), [Terraform](), [FastAPI](), [LangChain](), [Python]() Tags: [fastapi](), [github](), [persistence](), [pgvector](), [python](), [rag](), [retrieval](), [terraform]() ### AI overview The article explains why a RAG chatbot that works locally can lose vector indexes, conversation history, and documents when deployed across multiple replicas. It recommends shared persistence for embeddings, checkpointed conversation state, and shared object storage, with examples using Postgres, pgvector, LangGraph, and object storage. ### Source excerpt + a popular LLM interview question. ## Introducing agentic video understanding with Gemini DevFeed: [Introducing agentic video understanding with Gemini]() Original publisher: [Read original article]() Author: Rohan Doshi Published: 2026-09-01T17:08:51Z Content type: release Language: en Sources: [Google DeepMind News]() Topics: [Google AI](), [Google]() Tags: [agentic](), [ai](), [analysis](), [api](), [audio](), [benchmarks](), [cost](), [developers](), [enterprise](), [feature](), [flash](), [gemini](), [google-ai](), [models](), [none](), [performance](), [reasoning](), [retrieval](), [tools](), [transcripts](), [video]() ### AI overview Google launches agentic video understanding for Gemini Flash models. It uses native video tools to inspect relevant frames, audio, and transcripts, aiming to improve video-analysis accuracy while reducing token use and cost. ### Source excerpt We're launching agentic video understanding across our latest Gemini models for improved accuracy and lower costs and token usage. ## KV vs Prefix vs Prompt vs Semantic Caching DevFeed: [KV vs Prefix vs Prompt vs Semantic Caching]() Original publisher: [Read original article]() Author: Avi Chawla Published: 2026-08-27T20:01:47Z Content type: article Language: en Sources: [Daily Dose of Data Science]() Topics: [Caching](), [Large Language Model](), [Neo4j]() Tags: [caching](), [embedding](), [knowledge-graph](), [latency](), [llm](), [memory](), [nvme](), [object-storage](), [retrieval](), [server](), [token](), [vector]() ### AI overview The article distinguishes KV caching, prefix caching, prompt caching, and semantic caching in LLM systems. It explains what each cache stores, how exact-match and fuzzy-match behavior differ, and how cache misses affect cost, latency, and correctness. It also describes Konig's tiered storage approach for many small knowledge graphs, though the supplied text is truncated. ### Source excerpt ...explained with best practices in production. ## When LLM judges agree, should we believe them? DevFeed: [When LLM judges agree, should we believe them?]() Original publisher: [Read original article]() Author: Krishna Balasubramanian; Sasha Podkopaev Published: 2026-08-26T17:10:40Z Content type: article Language: en Sources: [Amazon Science homepage]() Topics: [Large Language Model](), [Retrieval-Augmented Generation](), [Ising](), [benchmark overfitting machine learning](), [Network](), [AI, ML & Data Engineering]() Tags: [conversational-ai](), [evaluation](), [ising](), [large-language-models-llms](), [llm](), [machine-learning](), [metrics](), [retrieval](), [retrieval-augmented-generation]() ### AI overview The article examines whether agreement among LLM judges is trustworthy when their outputs may be correlated. It presents a dependence-aware aggregation method based on Ising models that discounts shared blind spots and outperforms historical-accuracy-weighted majority voting on three tasks. ### Source excerpt Discounting the opinions of LLM judges with highly correlated outputs ensures that panels of judges reflect a true diversity of perspectives. ## Scaling Conditional Learned Retrieval for Pinterest Home Feed DevFeed: [Scaling Conditional Learned Retrieval for Pinterest Home Feed]() Original publisher: [Read original article]() Author: Pinterest Engineering Published: 2026-08-26T14:01:05Z Content type: article Language: en Sources: [Pinterest Engineering Blog - Medium]() Topics: [recommendation systems](), [Embeddings](), [Machine Learning & Artificial Intelligence]() Tags: [blog](), [embedding](), [eng-culture](), [engineering](), [infrastructure](), [pinner-experience](), [pinterest](), [ranking](), [retrieval]() ### AI overview Pinterest describes scaling Conditional Learned Retrieval for home-feed candidate generation. The approach extends two-tower retrieval with explicit contexts that produce multiple user embeddings for different interests and retrieval use cases. ### Source excerpt Devin Kreuzer | Sr. Machine Learning Engineer; Yichi Wang | Machine Learning Engineer I; Sujan Reddy Ale | Machine Learning Engineer I; Zelun Wang | Sr. Machine Learning Engineer; Hongtao Lin | Sr. Machine Learning Engineer; Piyush Maheshwari | Staff Machine Learning Engineer Pinterest home feed candidate generation is a large-scale User-to-Pin retrieval problem. A common approach is a two-tower model: a user tower encodes the user, an item tower encodes candidate Pins, and approximate nearest neighbor search retrieves Pins close to the user embedding. But Pinterest users often have multiple intentions at once -- planning a renovation, saving recipes, exploring fashion, or organizing travel ideas. A single retrieval embedding can struggle to capture this diversity. Conditional Learned Retrieval, or CLR, extends the two-tower setup by conditioning the user tower on an explicit retrieval context. Instead of producing only one user embedding, CLR can generate condition-aware embeddings that reflect different aspects of a user's interests while still grounding retrieval in the user's overall behavior. Prior Pinterest work studied this formulation in two settings. The RecSys'24 paper: Bootstrapping Conditional Retrieval for User-to-Item Recommendations described how to bootstrap conditional retrieval by constructing training data for (user, condition) -> item retrieval from existing user-item and condition signals, and applied it to interest-based notifications. The KDD'25 paper: Synergizing Implicit and Explicit User Interests: A Multi-Embedding Retrieval Framework at Pinterest placed Conditional Retrieval within a broader multi-embedding retrieval framework for home feed, where explicit interest conditions complement implicit interests extracted from user behavior. In this blog, we describe how CLR evolved from early interest-conditioned retrieval into a broader retrieval system for Pinterest home feed. We focus on three areas: expanding CLR to support more retrieval us ## Training and Finetuning Multi-Vector Embedding Models with Sentence Transformers DevFeed: [Training and Finetuning Multi-Vector Embedding Models with Sentence Transformers]() Original publisher: [Read original article]() Author: Tom Aarsen Published: 2026-08-26T00:00:00Z Content type: tutorial Language: en Sources: [Hugging Face - Blog]() Topics: [sentence-transformers](), [AI search](), [AI, ML & Data Engineering]() Tags: [community](), [datasets](), [embedding](), [embeddings](), [evaluation](), [guide](), [nlp](), [open-source](), [retrieval](), [sentence-transformers](), [training]() ### AI overview A tutorial on finetuning multi-vector embedding models with Sentence Transformers. It explains late-interaction token-level retrieval, training components, and domain-specific retrieval improvements. ### Source excerpt We're on a journey to advance and democratize artificial intelligence through open source and open science. ## Inside LinkedIn's cognitive memory agent for agentic personalization DevFeed: [Inside LinkedIn's cognitive memory agent for agentic personalization]() Original publisher: [Read original article]() Author: Phoebe Sajor Published: 2026-08-25T07:40:00Z Content type: article Language: en Sources: [Stack Overflow Blog]() Topics: [Artificial Intelligence](), [personalization](), [Latency](), [Authorization](), [data](), [Software]() Tags: [agent](), [agentic](), [agentic-ai](), [ai](), [latency](), [linkedin](), [personalization](), [podcast](), [retrieval](), [se-stackoverflow](), [se-tech]() ### AI overview This podcast discusses LinkedIn's cognitive memory agent, a four-layer memory system designed to give its hiring assistant persistent, personalized state. The conversation covers the move from GraphRAG to tree-structured memory, incremental updates, retrieval freshness, latency budgets, and access control. ### Source excerpt Ryan is joined by Praveen Bodigutla, Principal AI Researcher at LinkedIn, to chat about the four-layer memory system his team built to give LinkedIn's hiring assistant a persistent, personalized state. [Next page]()