Skip to content
Back to homeha.
✦ Selected work

Projects & case studies

Real engineering problems, measurable outcomes, and production results — spanning AI research, ML systems, and full-stack. Each entry has a deeper case study explaining the architecture and the decisions behind it.

MEAN CALL LATENCYn = 2,100+ concurrent stateful sessionsBeforeAfter0s0.6s1.2s1.8s2.4s2.4s1.1sREDUCTION−54% · 1.3sCONCURRENCY2,100+ sessions
★ Featured · In Production

Autonomous Voice Agent

2,100+ concurrent AI sales calls at 1.1s latency

Built at Outlyst — FastAPI inference backend for a Retell AI voice agent. Profiled with py-spy to isolate event-loop blocking, restructured asyncpg connection pools, and parallelised tool calls with asyncio.gather(). 2,100+ concurrent stateful sessions.

54%Latency reduction
2,100+Concurrent sessions
25%Lead conversion lift
FastAPIRetell AIAsyncIOPostgreSQL
Read full case study
Projects | Hammad Ahmad