KisaanRaksha
An early warning for farmer financial distress, in Marathi on WhatsApp
- Status
- 3rd prize · Hack4Humanity 2026
- Year
- 2026
- Rooms
- ML & Impact, AI & Agents
Problem
Support for distressed farmers in Vidarbha and Marathwada arrives after the crisis. The warning signs — drought, prices, crop health, loan dates — are public data no one combines in time.
Approach
A Financial Stress Index per district from four grounded signals, a Marathi voice agent on WhatsApp whose every fact comes from an MCP tool, and an officer dashboard that raises an alert above a critical threshold.
Engineering decisions
- No bare LLM calls: weather, mandi prices, NDVI and the FSI score all come from MCP tools backed by named datasets.
- Offline-first fallbacks for every signal when connectivity drops.
- A Fairlearn audit checks error parity across regions and crop groups.
What it does
- LightGBM Financial Stress Index
- Marathi voice via WhatsApp
- Officer alerts above FSI 75
- District heatmap dashboard
Stack
My part
Built for Hack4Humanity 2026 (Track C, AI for Societal Good), where it won 3rd prize; all 23 commits in the repository are mine.
Limits, stated plainly
- The FSI model is trained on a synthetic dataset built from public signals; the WhatsApp alert shown is a simulation replaying a past pattern.
- Reported metrics (MAE 3.75, R² 0.94) are on held-out talukas of that synthetic data.
Evidence
- Kisan-Raksha-project README
- model/training_metrics.json
- model/fairlearn_report.md
- Hack4Humanity 2026 3rd-prize certificate
Screenshots
A Marathi WhatsApp conversation showing a simulated critical FSI alert for Amravati