Back to projects

Prosecuto

May 2026 — Jun 2026

Prosecuto was built at NVIDIA Spark Hack Toronto on GB10 hardware. I owned the local-inference path: routing generation to an on-device llama-server and vLLM NVFP4 rather than a hosted API, and fixing a Redis lock that was dropping WebSocket connections on long turns. The stack is Python with LangGraph for orchestration, Chroma for retrieval, and a Next.js frontend.

Stack & Tools

PythonLangGraphvLLMChromaNext.jsLocal Inference

Links

Other Projects