NOTES

Occasional writing on Python, Kubernetes and machine learning in production.

  • 12 AUG 2026 · 1 MIN READ

    Rollback matters more than a perfect deploy

    The perfect deploy happens once a year; every system falls over eventually. That is why the first scenario our CI tests is not the rollout but the rollback.

  • 04 JUN 2026 · 1 MIN READ

    Models live in production, not on a laptop

    A laptop with 97% accuracy and a service running the same checkpoint are two different models. What changes when inference moves into Kubernetes.

  • 19 MAR 2026 · 1 MIN READ

    The boring layer in front of the model

    Half of all "inference problems" are solved by an nginx in front of FastAPI: limits, timeouts, caching and honest 503s. Cheaper than scaling GPUs.

COPIED