Guides
Field notes on shipping applied AI
Practical writing on what works and what breaks when AI features hit production. Aimed at engineering teams already in the weeds: less marketing, more diffs.
Nothing published yet. The first guides land here as engagements wrap and the writing finds its shape.
On the writing desk
- RAG
Why hybrid retrieval still beats pure vector search
- Evals
Evals as test suites: the minimum harness for an LLM feature
- Operations
Pricing AI features when token cost is non-deterministic
- Retrieval
When a cross-encoder reranker is worth the latency
- Agents
Tool-calling agents that don't loop forever