Degdeg — AI Climate-Health Early Warning

AI Climate-Health Early Warning · Somali Region, Ethiopia

Turning open climate forecasts into anticipated health impacts — before disaster becomes an emergency.

Plain-Somali advisories for woreda health workers — every recommendation cites a protocol rule and is approved by a person before it publishes.

View Live Demo Read Our Methodology

View live dashboard →

100 woredas monitored

Forecasts updated 10h ago

8 WHO & national protocol sources

~2,000 indexed protocol passages

Every advisory human-approved

The Problem

Early warning science is strong — but three gaps keep it from reaching the people who need to act:

The last-mile gap

Forecasts rarely reach woreda health offices or health workers in a form they can act on, in Somali.

The climate → health gap

A rainfall anomaly map doesn't tell a health officer to expect a cholera risk window in flood-hit kebeles.

The warning → action gap

Even when warnings arrive, nothing converts them into a protocol-based preparedness checklist.

How It Works

Four stages, each with an explicit human-auditability guarantee — rule citations and human approval are never skipped.

1. Interpret

The AI reads raw forecast and bulletin data and produces a graded hazard assessment: level, confidence, and its stated reasoning.

2. Anticipate

The assessment is checked against 15 vetted climate → health impact rules. Every anticipated risk must cite a matching rule — uncited output is rejected, not published.

3. Advise

The AI drafts advisory text — a WhatsApp message, radio script, and official memo — grounded only in the cited risks above, in Somali and English.

4. Approve

Nothing publishes automatically. A health officer or admin reviews and approves every advisory before it reaches a woreda.

Why You Can Trust It

"AI-native" doesn't mean unaccountable. Four non-negotiable rules keep the pipeline auditable end to end:

Grounding, not freestyling

Every anticipated health risk must cite a vetted impact rule, and its timing must fall inside that rule's range — unsupported output is rejected outright.

Structured validation

Every AI output is validated against a strict schema. Invalid output is retried once, then escalated to human review — never silently published.

Confidence + human-in-the-loop

Low-confidence or emergency-level assessments are automatically held back from public display until a human confirms them.

Full audit logging

Every AI call — prompt, model, response, validation outcome — is logged. That log is the methodology annex for due diligence.

Built for Funders and Partners

Degdeg is built for a conversation with funders like UNICEF, Wellcome, and Somali Region Health Bureau officials — a system that is accountable, auditable, and verifiable, not just another AI demo.

Read Our Methodology