Writing
Notes from building careful AI.
Long-form notes on production AI: evidence, evaluation, authority, and the engineering discipline around stochastic systems.
What 15 Years of Software Engineering Taught Me About AI Engineering
Which classical engineering disciplines survive contact with stochastic systems, and which problems are genuinely new.
AI engineering · software engineering · career · production AI
Building Reliable Agentic AI Systems
Agent reliability is an authority problem: what is permitted, for whom, until when, and what record survives the action.The Silent Zero: Proving Absence in Very Large Data
Zero rows is not an answer. What it takes to claim absence, and the wider family of claims that lean on a denominator.Designing Evidence Gates for Production LLM Systems
Models propose; gates certify. The five checks that stand between model output and effect, and the order they run in.How I Evaluate Production RAG Systems
Scoring the decision chain instead of the prose: retrieval, sufficiency, faithfulness, refusal quality, and honest denominators.Building AI That Cites or Refuses
The contract that makes AI auditable: every claim carries its records, or the system refuses and names the gap.