A collection of things I've built from small utilities to full applications.

Relay is a distributed LLM inference engine that runs models too large for a single machine by splitting transformer layers across pooled GPU/CPU nodes with pipeline parallelism, delivering token-for-token identical output to single-machine inference.
A real-time multi-camera AI surveillance platform where visitors are tracked across feeds via YOLO11 and OSNet Re-ID, projecting their live movement onto a 2D floor plan while flagging unauthorized zone entry and path anomalies with instant alerts.