Backend systems, applied ML, developer tooling — I build the thing when I can't find one that already does what I need.
Selected work
- rate-limiter — distributed multi-tenant limiter, atomic Redis Lua scripts, circuit-breaker fallback
- task-queue — priority task queue, gRPC + Postgres, dead-worker detection and DLQ
- emotion_recognition — speech emotion recognition, 95% multi-class accuracy, model to UI
More case studies → manel.vercel.app .


