A multimodal AI healthcare assistant for the Ethiopian context.
# ጤናለአዳም (Tena LeAdam) — Ethio-HealthBridge
**A multimodal AI healthcare assistant for the Ethiopian context.**
ጤናለአዳም (*Tena LeAdam* — "Health for Humanity") is an AI-powered triage and
medical-guidance system that accepts **Amharic voice, text, and images** and
responds in natural Amharic speech. Its answers are grounded in official
**Ethiopian Ministry of Health (MoH) clinical guidelines** via a retrieval-augmented
generation (RAG) pipeline. Patients can also escalate any conversation to a
hospital doctors' Telegram group on demand — either automatically when a
red-flag symptom is detected, or manually by tapping *ሐኪም ያማክሩ*
("Consult a doctor"). Reported symptoms are also aggregated per Addis Ababa
sub-city and surfaced on an MoH admin dashboard for outbreak hotspot
tracking.
> ⚠️ **Medical disclaimer.** This project is a research/hackathon prototype. It
> is **not** a substitute for professional medical advice, diagnosis, or
> treatment. In a real emergency, call **912** (Ethiopia) or your local
> emergency number.
---
## Table of Contents
- Key Features
- Architecture
- Tech Stack
- Repository Layout
- Prerequisites
- Quick Start
- 1. Clone the repository
- 2. Configure environment variables
- 3. Set up the backend
- 4. Ingest MoH guidelines into the vector database
- 5. Run the backend
- 6. Run the frontend
- Environment Variables
- Setting Up the Doctors' Telegram Group
- API Reference
- Scripts and Utilities
- Troubleshooting
- Contributing
- License
- Acknowledgements
---
## Key Features
- **Amharic Voice-to-Voice chat.** Speak in Amharic; the system transcribes,
reasons over MoH guidelines, and replies with a natural Amharic neural voice.
- **Grounded Medical RAG.** Answers are retrieved from official MoH Clinical
Guidelines stored in a persistent ChromaDB vector store, then summarized by
Gemini under a strict Amharic-only doctor persona.
- **Medicine Label Scanner (Vision).** Upload a photo of a drug label and get
an Amharic breakdown of name, do …