Village Health Data Assistant

Inspiration

– In rural South India, ASHA (Accredited Social Health Activist) workers collect health data on paper. – These reports take days or weeks to reach district health offices, delaying outbreak response. – We wanted to enable instant, structured, multilingual health reporting to detect diseases before they spread.

What it does

– Accepts voice or text inputs in Malayalam and Tamil from ASHA workers. – Converts inputs into structured health records (village name, disease type, case count, medicine stock). – Performs real-time threshold checks to identify abnormal spikes. – Sends instant alerts to health officers via SMS or dashboard notifications. – Eliminates delays and helps teams act within hours instead of days.

How we built it

– Speech-to-text transcription using IBM Granite models for Malayalam and Tamil. – Data extraction and parsing with ADK workflows to generate structured outputs. – Alert logic to compare case counts against preset thresholds. – Prototype outputs as CSV files with a dashboard for visualization.

Challenges we ran into

– Achieving high transcription accuracy for regional language variations. – Cleaning and structuring noisy voice inputs from rural settings. – Designing a lightweight system for low-connectivity environments.

Accomplishments that we're proud of

– Built a multilingual working prototype in just a few days. – Successfully tested real-time outbreak alerts with sample data. – Created a solution that can scale nationwide and integrate with health systems.

What we learned

– Voice-first rural reporting works — ASHA workers can use it easily. – Language inclusivity matters — supporting Malayalam and Tamil built immediate trust. – AI can reduce reporting time and human error significantly.

What's next for Village Health Data Assistant

– Add more Indian languages (Kannada, Telugu, Odia). – Integrate with state health dashboards for live monitoring. – Run field pilots with ASHA workers to validate accuracy. – Improve offline-first features for poor connectivity areas.

Built With

• adk
• granite
• ibm-watson
• python
• streamlit