Jan 2025 – Mar 2025
CerebrumAI
Offline-first multimodal healthcare triage engine with quantized LLMs, deployable on a Jetson Nano.
What I built
- Built a privacy-first triage system for early detection of mental and cognitive disorders from multimodal inputs (text, image, voice).
- Modular architecture with quantized LLMs (Phi-2, Bio-Medical LLaMA-3-8B) and runtime model switching.
- Offline-first, on-device inference via Llama.cpp on low-resource hardware (Jetson Nano, laptops, Linux servers).
- LangChain backend with prompt routing and ChromaDB vector search; Supabase for auth and data; Flutter and React frontends.
Stack
FlaskLlama.cppChromaDBPyTorchFlutterSupabase