all work

High-frequency trading matching engine in C++17

Architected a high-performance engine with ~230µs latency, async WAL, WebSocket thread pool for L2 market data, trailing stops, stop-loss, and crash-recovery from 100k+ log entries.

AI Technical Interviewer — full-duplex voice agent

Engineered voice agent with FastAPI WebSockets, Deepgram Nova-2, ElevenLabs (<800ms latency). Stateful DSA assessment with Gemini 2.5, Redis context, secure sandbox, and 99% session continuity.

High-Performance AI Gateway

Engineered a scalable AI Gateway and published a modular Go package for rate-limiting and cost-tracking middleware, cutting redundant upstream API calls by 40% via semantic caching. Secured 99.9% system availability with a fault-tolerant resilience layer (circuit breakers, Prometheus monitoring), designed as a reusable library for high-concurrency applications.