Offline AI-Powered Medical Translator (MVP)
Not medical advice. For emergency communication assistance only.

MediTongue is an offline medical translator that runs entirely on your local machine — no internet required.
It uses the open-source gpt-oss-20b model to translate medical dialogue between languages and detect urgent symptoms like chest pain or severe breathing issues.

Built for field healthcare, rural clinics, and low-connectivity environments.

• Offline AI translations — works without internet once the model is downloaded.
• Emergency flagging — instantly detects urgent symptoms and alerts the user.
• Medical term glossary — highlights and explains key medical terms.
• Quick-access phrasebook — preloaded with common emergency phrases.
• Cross-platform — runs locally on macOS, Windows, and Linux with Ollama.

• Frontend: Next.js, TailwindCSS, shadcn/ui
• Backend API: Node.js + Express
• LLM Runtime: Ollama (local)
• Model: gpt-oss-20b
• Emergency detection: Keyword spotting + AI classification

Download & install Ollama from:
👉 https://ollama.com/download

ollama pull gpt-oss:20b

git clone https://github.com/YOUR_USERNAME/meditongue.git
cd meditongue

# API
cd meditongue-api
npm install

# Web UI
cd ../meditongue-web
npm install

cd meditongue-api
export MODEL_BACKEND=ollama
export OLLAMA_MODEL=gpt-oss:20b
npm start

ollama serve

cd meditongue-web
npm run dev

Then open http://localhost:3000 in your browser.

Input (English):

The patient is having chest pain and shortness of breath.

Output (Spanish):

Tengo dolor torácico y dificultad respiratoria.
Flags: ["EMERGENCY"]
Terms:
- chest pain → dolor torácico
- shortness of breath → dificultad respiratoria

1. User Input → Typed into the UI.
2. Backend → Sends text to Ollama with the gpt-oss-20b model.
3. Translation → AI translates & checks for emergency keywords.
4. Glossary → Matches and highlights medical terms.
5. Output → Translation + emergency flags + glossary shown in the UI.

This project is licensed under the Apache 2.0 License.

• gpt-oss-20b for the open-source model
• Ollama for local LLM runtime
• Hackathon organizers for the challenge

This project was submitted for the For Humanity and Best Local Agent categories.