The reading list
Essays, tech notes, and the in-between.
- AI·3 min read
design.md for an MUI Codebase: A Concrete Template
Part 4 was the meta-template. This is the full design.md for an MUI v6+ codebase — every token, variant, and rule, ready to copy.
- AI·3 min read
Session-Start Hooks That Pay for Themselves
Zero ongoing cost. Measurable improvement on every interaction. The underused feature of modern coding agents — what to put in it, and what not to.
- AI·3 min read
Writing Task-Specific Agent Prompts That Work First Try
If design.md is the agent's memory, task prompts are its muscle memory — the highest-leverage agentic asset you can build, and the most underused.
- AI·3 min read
What to Put in design.md: A Complete Template
A good design.md is a decision document, not a style guide. If a senior designer left tomorrow, would the next person make the same calls?
- AI·3 min read
Rules That Agents Actually Follow: Enforcement Over Aspiration
A perfectly written CLAUDE.md cannot fix a contradictory codebase. The most effective rules file is short, opinionated, and backed by tooling.
- AI·3 min read
A 3-Week Plan to Make Your Legacy React Codebase Agent-Ready
Three weeks, one engineer, one opinionated order. Stop the drift, collapse the duplicates, and let the rules you already have actually start working.
- AI·3 min read
Why Your Legacy React Codebase Confuses AI Coding Agents
AI agents stumble on mature codebases because they have accreted multiple valid answers to the same question — and an agent has no way to know which one is current.
- technical-notes·3 min read
Entity resolution for guests: fuzzy matching first, LLM disambiguation second
The same person shows up as 'Bibhusan Bista', 'Bibhusan B.', and 'B. Bista' across three episodes. Don't ask the LLM first — try cheap deterministic matching, then escalate only the ambiguous cases.
- Technical Notes·3 min read
Structured extraction with Pydantic + Claude: guests, topics, and quotes from raw transcripts
Schema-first prompting with Pydantic + Anthropic tool use, a Haiku triage pass that gates Sonnet extraction, prompt caching for the system block, and a single retry that feeds the validation error back into the prompt.
- technical-notes·3 min read
Ingesting YouTube transcripts: yt-dlp for subs, Whisper when subs don't exist
YouTube's Data API gives you channel discovery but not third-party caption text — that endpoint is OAuth-only in practice. The honest pipeline is yt-dlp in subs-only mode for captions, and local Whisper as a deeper fallback when no captions exist at all.
