Project
CivicGuard Africa
AI-assisted election disinformation triage and multilingual safety benchmarking for African contexts.
One-line pitch
CivicGuard Africa is a civic AI safety tool and evaluation framework for African election disinformation risks.
Problem
Election rumors, synthetic media claims, false public-service notices, and identity-targeted narratives can move faster than verification teams, especially through local-language and code-switched channels.
Solution
CivicGuard provides a structured local workflow for intake, explainable triage, community review, trend monitoring, and benchmark-driven AI safety evaluation. It is a hackathon MVP and research artifact, not a claim to production fact-checking.
Target users
Hackathon alignment
CivicGuard Africa targets the Africa track through two contribution types: Monitoring & Tooling, by providing a working civic-risk triage prototype, and Evaluation & Benchmarking, by providing a multilingual African election safety benchmark with exportable data and a written methodology.
Why it matters for Africa
African election information ecosystems are multilingual, mobile-first, and highly local. Civic-risk tools must account for local languages, community context, low-resource verification settings, and the reality that English-only AI safety testing can miss important harm patterns.
What was built in the MVP
- Incident submission and deterministic risk triage
- LocalStorage persistence for submitted incidents and reviewer notes
- Community review queue with status updates
- Risk dashboard with Recharts visual summaries
- Benchmark Lab for multilingual election-safety evaluation
- Exportable local incident and benchmark datasets
- Research documentation for methods, limitations, and submission framing
What comes next after the hackathon
- Connect Supabase for shared incident storage and role-based review workflows.
- Add authentication for civic partners, newsroom teams, and trusted reviewers.
- Integrate AI APIs for assisted classification while keeping human review in control.
- Partner with fact-checkers and election observers for localized evidence workflows.
- Expand benchmark coverage with more African languages, regions, and reviewer annotations.
Ethical safety note
CivicGuard does not determine truth automatically. It supports human verification by prioritising and structuring suspicious civic content. The MVP avoids real parties, real candidates, paid APIs, and operational manipulation examples.