MedoX

Empowering 1.2 million rural nurses with a multimodal AI assistant and real-time epidemic tracker powered by MedGemma.

Comment

Empowering 1.2 million rural nurses with a multimodal AI assistant and real-time epidemic tracker powered by MedGemma.

Inspiration

India has 1.2 million rural nurses wasting 3 hours a day on paper registers. Most health tech assumes they have a keyboard and stable internet. I wanted to build something that actually works at the last mile and gives them their time back.

What it does

It’s a zero-tap AI assistant. A nurse just speaks naturally about a patient. MedoX transcribes the audio, extracts a structured SBAR clinical note, and autonomously updates the local DB, a Google Sheets registry, and a public health dashboard. It also includes a clinical AI chat for quick second opinions.

How I built it

Frontend: React Native (Expo) for a simple, aesthetic UI.

Backend: Python / Django.

AI: Faster-Whisper for local transcription. The medical reasoning uses MedGemma-27b, accessed serverless via featherless.ai so it doesn't need local GPUs.

Automations: Google Sheets and Calendar APIs.

Challenges I ran into

Running a heavy medical AI in a low-resource environment is tough. I had to balance local, CPU-efficient transcription with remote API routing. Also, prompt engineering MedGemma to consistently return strict, structured JSON from messy, unstructured voice notes took a lot of trial and error.

Accomplishments that I'm proud of

The "zero-tap" pipeline. Watching a single voice note instantly transform into a structured record, update a spreadsheet, and schedule a calendar appointment without ever touching the screen is awesome.

What I learned

General LLMs hallucinate way too much for clinical use—you absolutely need a domain-specific model like MedGemma. I also learned that the biggest bottleneck in rural health isn't always medicine; it's just broken administrative infrastructure.

What's next for MedoX

Adding multi-language support (Hindi/Bengali) and building an SMS-fallback architecture, so if a clinic loses 4G completely, the app can still compress and text the structured data to the district server.

Built With

Share this project:

Updates