josuekb09/civicguard-africa

# CivicGuard Africa AI-assisted election disinformation triage and multilingual safety benchmarking for African contexts. GitHub repository: github.com Deployed demo: civic-guard-africa.vercel.a… ## Submission Framing CivicGuard Africa is a civic AI safety tool and evaluation framework for African election disinformation risks. This project targets the Global South AI Safety Hackathon Africa track through: - Monitoring & Tooling: a working prototype for suspicious election-content intake, deterministic triage, community review, and civic risk monitoring. - Evaluation & Benchmarking: a multilingual African election safety Benchmark Lab with safe prompt summaries, scoring labels, charts, and exportable datasets. The web app is the demo interface. The full submission also includes a research-style report, methodology, demo script, judge Q&A, and submission plan in the `docs/` folder. ## Problem Statement Election disinformation can spread faster than verification teams can respond. In African contexts, suspicious claims often move through WhatsApp groups, screenshots, voice notes, local languages, and code-switched community speech during high-pressure election periods. AI-generated or AI-amplified content can create risks around voter suppression, fake results, candidate impersonation, identity targeting, procedural misinformation, intimidation, synthetic media claims, and false public-service information. English-only AI safety checks are not enough for this environment. ## Solution CivicGuard Africa helps users: - Submit suspicious election-related content. - Generate a transparent deterministic triage report. - Simulate human/community review workflows. - Monitor civic risk patterns on a dashboard. - Evaluate AI safety behavior through a multilingual Benchmark Lab. - Export incident and benchmark data for future research. The MVP does not use Supabase, authentication, paid APIs, or external AI calls. It is …