MediTongue

Inspiration

In many parts of the world, doctors and patients face language barriers — sometimes in life-or-death situations. Internet-based translation tools fail in low-connectivity or offline settings, leaving healthcare providers without critical communication support. MediTongue was built to solve that.

What it does

MediTongue is an offline AI-powered medical translator that:

• Translates medical dialogue in real time between multiple languages
• Flags potential emergencies like chest pain or severe breathing issues
• Works completely offline, ensuring reliability anywhere
• Provides a quick-access emergency phrasebook for common urgent questions

How we built it

• Frontend: Next.js + TailwindCSS + shadcn/ui for a clean, responsive UI
• Backend: Node.js local API that communicates with the Ollama runtime
• Model: GPT-OSS 20B running locally for translations and symptom detection
• Emergency detection: Rule-based keyword spotting combined with AI classification
• Designed to gracefully handle offline mode, with health checks and latency badges

Challenges we ran into

• Optimizing large language model inference to run locally at acceptable speeds
• Ensuring translations remained medically accurate while still concise
• Designing a UI that works well on both desktop and mobile for field use
• Managing local model storage and startup times without overwhelming system resources

Accomplishments that we're proud of

• Achieved full offline functionality — no internet required for translations
• Emergency detection works instantly in both directions
• Built a simple, intuitive interface that non-technical users can pick up in seconds
• Integrated medical term highlighting for better clarity between doctor and patient

What we learned

• Offline AI is now powerful enough to serve real-world, high-stakes use cases
• Medical translations require a balance of precision, cultural sensitivity, and speed
• Even small UI tweaks (like flag emojis or phrasebook chips) can greatly improve usability

What's next for MediTongue

• Expand to more languages and dialects used in medical missions and rural healthcare
• Add voice input/output for completely hands-free translation
• Create a “triage mode” for paramedics and first responders
• Optimize model performance for low-power devices like tablets and rugged field laptops

Built With

• express.js
• gpt-oss-20b
• lucide-icons
• next.js
• node.js
• ollama
• react
• shadcn/ui
• sonner
• tailwindcss