Field notes
Patterns from production
Real problems from systems in production: what broke, why, how I fixed it and what changed.
- 01Shipping AI people can trustStructured prompts, scores that explain themselves, and output that reads naturally in Arabic and English.Read the case
- 02When the error message liesUsers were told their connection dropped. It was a server-side rejection all along. Now the UI tells the truth.Read the case
- 03Private by defaultRecordings opened for anyone who had the link. Now every request is checked before a single frame streams.Read the case
- 04A build on a dietRewrote the ignore rules so every deploy ships only what the image needs.Read the case
- 05When the database lost its primaryEvery node up, none accepting writes. Restored a healthy primary and wrote the runbook.Read the case
- 06Results that showed up too lateAsync results now land the moment they are ready, not at the end.Read the case
A thousand welcomes