ASAL Guardian: multi-agent drought warning
From field report to bilingual alert in minutes.
Three Gemini agents work in sequence: one structures a field report, one classifies the drought phase against NDMA-style thresholds and terms of trade, and one drafts a bilingual SMS alert and a brief to the county leadership. Built for the Google and Kaggle AI Agents Intensive capstone (Agents for Good).
The problem
Drought bulletins are compiled monthly, so response in pastoral areas often starts after losses begin.
What we built
- 1
Sentinel agent extracts vegetation condition, water distance and market prices as structured data.
- 2
Guardian agent classifies the drought phase and economic stress, with a stated reason.
- 3
Responder agent drafts an English and Kiswahili SMS and a one-paragraph brief.
- 4
A simple web view runs the whole chain with one click.
Highlights
- Clear, auditable thresholds rather than a black box
- Bilingual output sized for SMS
- Automatic fallback across available Gemini models
Built with
- Gemini
- Python
- Flask
- Multi-agent
Honest status: Prototype running on a simulated field report; no live data feeds are connected yet.
Our founder's capstone project.
Want something like this?
We design, build, host and secure systems like this end to end, and train your team to run them.
More success stories