CerebrumAI – Offline-First Multimodal Healthcare Triage Engine

• Built a privacy-first, AI-powered triage system enabling early detection of mental and cognitive disorders via multimodal inputs (text, image, behavior, voice).
• Designed modular architecture with quantized LLMs (Phi-2, Bio-Medical LLaMA-3-8B) and singleton pattern for real-time model switching and performance optimization.
• Implemented offline-first capabilities using Llama.cpp, ensuring HIPAA-aligned, on-device inference on low-resource systems (Jetson, MacBook, Linux servers).
• Developed a LangChain-powered backend with prompt routing, vector search (ChromaDB), and fault-tolerant endpoint monitoring with integrated alert hooks.
• Delivered secure user auth and data layer using Supabase; optimized frontends across platforms (Flutter and React) for responsive healthcare-grade UX.
• Engineered for scalability—from individual users to clinics and public health deployment—using a modular "Service-as-Block" system with native CI/CD and deployment scripts.