# backend-engineering Published articles for backend-engineering. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## How engineers at Nextdoor use Codex to build without limits DevFeed: [How engineers at Nextdoor use Codex to build without limits]() Original publisher: [Read original article]() Published: 2026-06-09T12:00:00Z Content type: article Language: en Sources: [OpenAI News]() Topics: [codex](), [developer-productivity](), [Product Management]() Tags: [about](), [accelerated](), [agent](), [alerts](), [analysis](), [away](), [backend](), [backend-engineering](), [backlog](), [better](), [bottlenecks](), [brand](), [build](), [building](), [but](), [codex](), [gpt-5-5](), [platforms](), [product](), [with]() ### AI overview Nextdoor engineers use Codex to work toward product outcomes across platforms and investigate difficult technical issues. In the Opportunity Alerts example, one engineer built a map feature end to end that would previously have involved mobile, frontend, and backend teams. For debugging, the team gives Codex a clean environment and harness to investigate problems such as Kubernetes pods failing to start and race-condition issues. Cory Dolphin says GPT-5.4 and GPT-5.5 have improved Codex's persistence in tracing root causes, while Fast Mode enables quicker feedback. He reports that engineering speed has shifted the bottleneck toward deciding what to build and setting product strategy. ### Source excerpt How engineers at Nextdoor use Codex with GPT-5.5 to investigate hard-to-reproduce issues, build across platforms, and focus on product outcomes. ## How Conduit's Early Convex Choice Shaped Its Engineering and AI Coding Practices DevFeed: [How Conduit's Early Convex Choice Shaped Its Engineering and AI Coding Practices]() Original publisher: [Read original article]() Author: Stack Published: 2026-03-18T17:36:18Z Content type: article Language: en Sources: [Stack]() Topics: [build vs buy database solutions](), [engineering-leadership](), [human review](), [Single-page application (SPA)](), [Publish-subscribe pattern](), [NUMA](), [cache-invalidation](), [Image]() Tags: [ai](), [ai-coding-tools](), [ai-generated-code](), [backend](), [backend-engineering](), [clients](), [coding](), [concurrency](), [convex](), [production](), [real-time](), [scale](), [ui](), [websockets]() ### AI overview Conduit adopted Convex after a hackathon and kept using it as the company grew. The article describes how its real-time backend, explicit query and pagination patterns, and collaboration with Convex influenced the team's shipping cadence, onboarding, and engineering practices. It also explains how established codebase patterns helped guide AI-generated code, while noting remaining platform limitations. ### Source excerpt This video explores how a Y Combinator-backed startup transitioned from a TreeHacks hackathon project to a production-scale real-time application using Convex. The discussion covers the architectural decisions behind choosing a managed backend, highlighting the immediate benefits of built-in WebSockets for UI reactivity and rapid iteration. It also delves into the early technical hurdles of crossing the Node and V8 chasm when handling specialized workloads. ## Plugging Node.js Memory Leaks: Optimisation Insights DevFeed: [Plugging Node.js Memory Leaks: Optimisation Insights]() Original publisher: [Read original article]() Author: Sowmiya Devarajan Published: 2026-02-25T08:31:35Z Content type: news Language: en Sources: [Flipkart Tech Blog]() Topics: [Memory Leaks](), [observability](), [Distributed Systems & Parallel Computing]() Tags: [advertising](), [backend-engineering](), [frontend](), [graphql](), [leaks](), [memory](), [memory-leak](), [memory-leaks](), [node-js](), [nodejs](), [optimisation](), [performance-optimisation]() ### AI overview Flipkart's Brand Self Serve Node.js GraphQL service became unstable under sustained load around 300 queries per second, with rising CPU and memory, failed health checks, and out-of-memory crashes. The investigation identified logging overhead, request-context retention, and server-side rendering (SSR) cleanup as contributing issues. Upgrading Winston to 3.11.0 or later addressed memory and CPU overhead from buffered log data. Replacing full request/response objects in Continuation-Local Storage with lightweight identifiers reduced retention; migration to Node.js AsyncLocalStorage was described as a next step. The team also restricted SSR to shell components and moved to client-side rendering, reporting no noticeable page-load degradation; upgrading Apollo Client from 3.1 was presented as a future option. After the combined changes, CPU stayed below 20%, memory remained around 500-600 MB over multiple days, and heap growth after repeated navigations was limited to 11 MB. The article attributes the overall improvement to the combined optimizations rather than isolating each change's contribution. ### Source excerpt Brand Self Serve (BSS) is a core web application within Flipkart(https://advertising.flipkart.com/), designed for advertisers (Brands and Sellers) to create and manage advertisements in multiple formats. These ads are strategically placed on designated slots across flipkart.com pages, ensuring visibility and engagement with millions of shoppers. The platform offers advertisers a seamless, intuitive user interface, built with ReactJS on the frontend and powered by a Node.js GraphQL server on the backend. This architecture efficiently handles complex ad creation workflows, campaign management, and real-time API requests, making the system both flexible and scalable. Given its central role in helping advertisers reach customers, the reliability of this service is critical. During periods of sustained traffic, particularly when query rates spiked, the system experienced challenges that affected stability. These issues highlighted areas where advertiser workflows were interrupted, causing delays and disruptions that affected their ability to configure and manage ads smoothly. This blog describes how the issues were diagnosed, the patterns applied effectively, and the trade-offs considered to maintain performance without compromising system integrity. It also emphasizes the importance of continuously strengthening scalability and resilience in the service's backend, ensuring advertisers continue to have a smooth experience even during sustained traffic load. The performance results are presented at the overall level, highlighting the cumulative effect of all optimisations. Instead of isolating incremental changes, the narrative underscores the substantial transformation achieved through collective improvements. Optimisation Highlights Stability: Successfully eliminated system restarts caused by resource exhaustion. Efficiency: Reduced peak CPU load near total saturation to 20% and transitioned from persistent memory leaks to a lean 15% footprint. The Crisis Our Node.js Gr ## From command line to community impact: How Shalini Ananda built Fire Fairness with Lovable DevFeed: [From command line to community impact: How Shalini Ananda built Fire Fairness with Lovable]() Original publisher: [Read original article]() Author: Ash Oliver Published: 2025-09-03T17:35:00Z Content type: article Language: en Sources: [Lovable]() Topics: [cursor](), [UI/UX Design]() Tags: [2](), [ai](), [audit](), [backend](), [backend-engineering](), [detection](), [engineering](), [front-end](), [generate](), [geolocation](), [lovable](), [stories]() ### AI overview AI researcher Shalini Ananda used Lovable to turn backend code and AI agents into Fire Fairness, a web app for legal aid groups processing wildfire insurance claims. The article says the tool verifies whether addresses fall within burn impact areas, identifies demographic bias patterns, and validates damage using public fire data. It reports use by legal aid groups and faster claim processing, and describes plans to keep the tool secure and reliable as it scales. ### Source excerpt How Shalini Ananda created a fire equity engine to support victims of the 2025 Los Angeles wildfires by using Lovable. ## How we built ngrok's data platform DevFeed: [How we built ngrok's data platform]() Original publisher: [Read original article]() Author: Christian Hollinger Published: 2024-09-26T00:00:00Z Content type: article Language: en Sources: [\[ngrok news\]]() Topics: [data lake](), [Apache Kafka](), [AWS Glue](), [CI/CD]() Tags: [apache](), [apache-flink](), [apache-iceberg](), [architecture](), [aws-glue](), [backend-engineering](), [bazel](), [company](), [data-lake](), [engineering]() ### AI overview Ngrok describes how a small engineering team built and evolved its data platform, including batch ingestion, streaming Protobuf events through Kafka and Flink into Iceberg, and analytics with dbt and Superset. The article also details schema handling, integrating data tools with the Go monorepo and CI, and using platform data for abuse investigations. ### Source excerpt At ngrok, we manage a ~100TiB, 500+ table data lake, managed by a very small team. Here's a look at how we built it and what unique challenges we solved. ## How We Horizontally Scaled Function Execution DevFeed: [How We Horizontally Scaled Function Execution]() Original publisher: [Read original article]() Author: Stack Published: 2024-04-05T19:00:00Z Content type: article Language: en Sources: [Stack]() Topics: [scalable-architecture](), [Transactions](), [monolithic-architecture]() Tags: [backend-engineering](), [caching](), [concurrency](), [convex](), [perspectives]() ### AI overview Convex separated function execution from customer backend processes into Funrun, a horizontally scalable, multi-tenant service. Backends retain transaction conflict checks and commits, while Funrun executes functions against database snapshots. Caching, consistent request routing, and isolate scheduling helped preserve performance, with reported median execution times below 20 milliseconds. The rollout enabled pro customers to run ten times as many functions concurrently. ### Source excerpt We shipped a behind-the-scenes update that lets customers run 10x concurrent functions than they used to. Emma, who helped ship the update, provides the context, architecture, and impact. ## How to build connected travel apps with Apollo GraphOS DevFeed: [How to build connected travel apps with Apollo GraphOS]() Original publisher: [Read original article]() Author: Shane Myrick Published: 2023-07-14T13:21:25Z Content type: article Language: en Sources: [Apollo Blog]() Topics: [GraphOS]() Tags: [api](), [apollo](), [architecture](), [attacks](), [backend](), [backend-engineering](), [best-practices](), [bot](), [connect](), [devices](), [e-commerce](), [faster](), [features](), [financial](), [frontend](), [graphos](), [graphql](), [technology](), [travel]() ### AI overview A travel supergraph gives frontend developers a consistent API for retrieving data across devices and building features as backend teams expose capabilities in the schema. The source says this can help product teams ship faster while allowing backend teams to manage relevant schema areas independently, with safety from Apollo GraphOS. It points readers to an explorable schema and real queries, plus guidance on SDUI, partner integrations, personalization, and mitigating scraping and bot attacks. ### Source excerpt Travel companies like Expedia Group, Booking.com, and Priceline have unique opportunities to connect their customers with the rest of the world. Unlike other consumer industries, travel companies are not just another e-commerce store; They provide opportunities to customers that we all wish we had more time in life to experience, and unique opportunities require unique technology solutions. ## Technical Interview Reform, Part 1: Rethinking the Backend Engineering Interview Take-Home Challenge DevFeed: [Technical Interview Reform, Part 1: Rethinking the Backend Engineering Interview Take-Home Challenge]() Original publisher: [Read original article]() Published: 2020-06-25T00:00:00Z Content type: article Language: en Sources: [SoundCloud Backstage]() Topics: [Software Engineering](), [Back end](), [engineering-culture](), [Programming](), [Test-driven development]() Tags: [academia](), [algorithms](), [algorithms-and-data-structures](), [backend](), [backend-engineering](), [bias](), [browser](), [candidate](), [challenge](), [code](), [code-challenge](), [code-review](), [code-reviews](), [coding](), [coding-environment](), [data-structures](), [engineering](), [interview](), [interviewing](), [technical](), [technical-interview]() ### AI overview SoundCloud reworked its backend engineering take-home challenge after candidates reported spending 12 hours or more over multiple weeks, and hiring data showed substantial dropoff at that stage. The revised exercise keeps the same coding problem but asks candidates to refactor working code while preserving a passing tester, then add a small feature. Candidates choose from six languages and use only the chosen language's standard library; clearer instructions spell out expectations. SoundCloud says the format retains an anonymous, more focused evaluation while reducing candidate effort. Two reviewers assess each submission independently, then agree on a consensus review; shorter reviews make it easier to add tiebreakers and respond sooner. A small pull request challenge also lets candidates demonstrate code critique. The excerpt gives no outcome data; it points to part 2 for quantitative and qualitative results. ### Source excerpt Most SoundCloud backend engineers have good feelings about the old backend engineering take-home challenge. It's commonly been characterized...