Field Notes

Writing

Working notes on Go infrastructure, RAG pipelines, Claude Code agents, evaluation, and the systems thinking behind software that has to keep working.

  1. 01

    From LBJ to FHIR: What J-codes taught me about US healthcare coding

    A builder's tour of why drug claims get denied, what the two-tier HCPCS system actually is, and why the 2026 FHIR mandate quietly carved out drug prior authorization.

    healthcarefhirx12hcpcsinteroperability
  2. 02

    Running Claude Code on a Hetzner VPS

    How I run Claude Code on a Hetzner VPS over Tailscale: the components I use, the setup process, and an automated installer that compresses a 90-minute walkthrough into roughly ten.

    claude-codevpstailscalehetznerremote-development
  3. 03

    Sandboxing Claude Code in Docker: From Naive to Hardened

    How I evolved a containerized Claude Code setup from 'it works' to actually secure — Docker secrets, read-only mounts, scratch clones, non-root runtime, and what Anthropic's own reference does differently.

    claude-codedockersecurityai-agentsdevcontainer
  4. 04

    Reliable AI Writing Systems

    A research proposal for detecting and reducing hallucinations in long-form AI writing — problem decomposition, evaluation rubric, proposed architecture, and validation plan.

    aihallucinationssystem-designevaluation
  5. 05

    A/B Testing Your RAG Pipeline: Chunking, Retrieval, and Reranking Strategies You Can Build With One Prompt Each

    How to quickly build and compare RAG pipeline variants — cosine vs. hybrid search, fixed vs. semantic chunking, Cohere vs. cross-encoder reranking — using Claude Code, Graphite stacks, and your own offline evals.

    ragclaude-codeai-agentsgraphiteevaluation
  6. 06

    Building Resilient Systems at Scale

    Lessons learned from scaling a legal-tech platform from prototype to processing millions in claims.

    goinfrastructurestartups
  7. 07

    Part 1: Data Pipelines in Golang

    Build a tiny, production-style pipeline in Go that crawls a public web page, transforms the result, and uploads it somewhere useful — all in ~150 lines of code.

    godata-pipelinesetl
← Back home