Blog

Practical notes from backend, data, and production AI work: reliability, testing, deployment, performance, and cloud costs.

• September 11, 2026

  The Ideal CI/CD Deployment Pipeline

  From build checks to production, how to structure a CI/CD pipeline that catches failures early and deploys changes safely.

• September 9, 2026

  5 Types of Software Testing for More Reliable Systems

  A practical guide to unit, integration, canary, fault injection, and performance tests, with lessons from almost a decade at Amazon.

• September 7, 2026

  Don't Manage Production Prompt Labels by Hand

  If your code is versioned but production prompts sit behind a mutable label, you've only versioned half the application.

• September 4, 2026

  Your Prompts Need Integration Tests Too

  A playground checks the prompt. Production failures often come from retrieval, parsing, and the code around it.

• September 3, 2026

  When Amazon EMR Gives YARN More Memory Than the Machine Has

  On some instance types, EMR tells YARN it can allocate more RAM than Linux sees. Spark looks correctly sized. The node still runs out of memory.

• September 2, 2026

  Why Your Spark Job Keeps OOMing Even After You Scale Amazon EMR

  EMR sizes executors for the hardware, not your workload. Scaling out adds more of the same undersized executors, and mixed instance fleets can make it worse.