Husika Alert

Inspiration

Regional institutions like ICPAC already produce excellent hazard and drought monitoring data through platforms like HUSIKA, Hazard Watch, and Drought Watch. But that data is built for technical analysis — indicators like VCI, SPI-3, and WRSI mean little to a farmer in Turkana or a family near the Shabelle river basin who just needs to know: is my area at risk, and what should I do right now? We wanted to build the "last mile" translation layer that turns expert-grade monitoring into something anyone can read and act on in seconds.

What it does

Husika Alert ingests hazard indicator data — drought signals, flood river-stage readings, food security phase classifications — and converts each one into a plain-language alert: a short headline, a 2-3 sentence explanation of the risk, and a concrete list of actions. Alerts are available in English and Swahili, and each one stays linked to the raw indicator it came from, so nothing is a black box.

How we built it

The app is a React single-page interface. Each hazard record is structured the way ICPAC's own indicators are: a value, a trigger threshold, and a status. A local rule-based engine renders instant plain-language alerts so the interface never stalls. Each card also connects live to the Claude API — raw indicator data is sent in a structured prompt, and Claude returns a JSON object with severity, headline, message, and action steps, demonstrating a genuine AI transformation layer rather than a static mockup.

Challenges we ran into

Keeping the demo reliable while still using live AI was the core tension — we solved it by treating the local templates as the dependable baseline and AI generation as a live enhancement layer on top. Getting Swahili translations right also mattered more than we first expected: a wrong word in a warning message isn't just an inconvenience, it could cost someone the time they need to act.

Accomplishments that we're proud of

We built a working demonstration of how AI can bridge the gap between advanced climate monitoring systems and the communities they're meant to protect — showing real, structured hazard data being turned into warnings a non-technical reader can understand and act on immediately.

What we learned

We learned how much design work goes into making "simple" language actually simple without losing the accuracy that makes a warning trustworthy — and how valuable it is to pair AI generation with a reliable fallback so a life-safety tool never goes silent.

What's next for Husika Alert

Connecting to ICPAC's live Hazard Watch, Drought Watch, and Thresholds & Triggers feeds directly, adding SMS/USSD delivery for low-connectivity areas, expanding language coverage across the IGAD region, and working with local disaster response coordinators to validate alert accuracy in the field.

Built With

• anthropic
• claude-api
• climate-data
• disaster
• early-warning-systems
• javascript
• json
• lucide-icons
• natural-language-generation
• react
• risk
• swahili
• tailwind-css