Writing
Field guides and build notes on AI that has to work in production: retrieval, orchestration, evaluation, and the systems underneath.
4 field guides · 8 build notes
Field guide12 min read
1.15×rank 1 vs rank 10 under RRF at k = 60
RAG past the tutorial: where retrieval quietly breaks
Field guide10 min read 1.7% vs 60%same pass@1, very different pass^8Agent orchestration, the hard parts
Field guide8 min read 200 → 54effective questions when they share documentsEvaluating RAG end to end without fooling yourself
Field guide8 min read 62.7 vs 99.9same model, two harnesses, ARC-AGI-3An AGI benchmark score is not a coding score
Build note3 min read 350+agents, one control planeMission control for 350 agents
Build note3 min read 78 → 864signals after the auditA 200 doesn’t mean your data is useful
Build note2 min read 756 → 28connections at 28 apps28 apps, 756 connections: why the database is the contract
Build note2 min read 9 stagesfrom task to contextWhy my context engine leaves budget unspent
Build note2 min read 300people matched, liveLet the code decide, let the model explain
Build note2 min read 395+round trips avoidedWhen not to use your own gateway
Build note2 min read 206live agents syncedSkills that get better by losing
Build note2 min read 7 stepslink to Arabic lectureDubbing Arabic without drifting off the speaker
Contact
Let's build
Building something like this, or want me to write about your problem next? One email is enough to start.