Health Heaven — Offline AI Health Assistant

🌱 Inspiration

During natural disasters, conflicts, or rural isolation, people often lack access to immediate medical guidance. Internet-dependent AI tools are unavailable in these moments, yet decisions in the first minutes can be life-saving. We wanted to build a system that works entirely offline, powered by reasoning models, to offer first aid, triage, and safety advice — no connectivity required.

🛠️ How We Built It

Core Model: We integrated OpenAI’s gpt-oss reasoning models for structured medical triage.
Deployment: Runs on low-power hardware like Raspberry Pi or laptops, ensuring accessibility.
Interfaces:
- CLI: Simple text-based interface for testing and offline usage.
- API: FastAPI server for local apps.
- Voice Agent: Speech-to-text with Vosk + speech synthesis with Coqui TTS.
Safety Layer: Prompts enforce disclaimers and structured reasoning output: $\text{Output} = \{ \text{triage}, \text{reasoning}, \text{safety advice} \}$

📚 What We Learned

Deploying large reasoning models offline requires careful optimization — quantization, adapter tuning, and prompt engineering.
Voice interaction dramatically improves accessibility in emergencies.
Community medical data (e.g., WHO, Red Cross first-aid manuals) can be safely adapted for fine-tuning.

🚧 Challenges

Model Size vs. Hardware: Running GPT-style models on Raspberry Pi required pruning, quantization, and efficient caching.
Speech Latency: Ensuring transcription + response within a few seconds for a smooth voice experience.
Safety Guardrails: Balancing helpfulness with strict disclaimers, so the system never replaces professional medical care.
Energy Constraints: Optimizing for battery-powered devices in field conditions.

🌍 Impact

Health Heaven has the potential to bridge healthcare gaps in underserved areas, empower first responders, and save lives during crises — all while keeping data private and offline.

Built With

built-with-languages:-python-?-frameworks-&-libraries:-transformers-(model-loading-&-inference)-torch-(deep-learning-backend)-fastapi-(local-api)-uvicorn-(asgi-server)-vosk-(speech-recognition)-coqui-tts-(text-to-speech)-platforms:-raspberry-pi
linux
macos
python
virtualenv
windows-containerization:-docker-?-other-tools:-hugging-face-hub-(model-weights)