# CivicGuard Africa Presentation Outline

## Slide 1: Title

**CivicGuard Africa**  
AI-assisted election disinformation triage and multilingual safety benchmarking for African contexts  
Josue Kabuya Tambwe

**Speaker notes:**  
Hi, I am Josue Kabuya Tambwe. My project is CivicGuard Africa, a civic AI safety tool and evaluation framework focused on election disinformation risks in African contexts.

## Slide 2: Problem

- Election misinformation can spread faster than verification teams can respond.
- In African contexts, this often moves through WhatsApp, social media, local languages, screenshots, voice notes, and code-switching.
- English-only AI safety testing can miss these risks.

**Speaker notes:**  
The problem is not only that misinformation exists. The problem is speed, language, and context. During elections, suspicious claims can move quickly across informal channels, while verification teams need time to check evidence.

## Slide 3: What I Built

A working MVP where users can:

- Submit suspicious election-related content
- Generate a transparent triage report
- Simulate human/community review
- Monitor risk trends on a dashboard
- Explore a multilingual Benchmark Lab

**Speaker notes:**  
CivicGuard is not just a landing page. The demo has a full workflow: incident intake, triage, review, dashboard monitoring, and a Benchmark Lab for evaluating AI safety behavior across African election contexts.

## Slide 4: How It Works

Submit incident -> deterministic triage -> reviewer checklist -> community review queue -> dashboard -> benchmark export.

**Important:** CivicGuard does not determine truth automatically. It supports human verification.

**Speaker notes:**  
The MVP uses deterministic triage instead of a paid AI API. That keeps the demo explainable and reliable. It gives reviewers a risk level, harm category, checklist, and next steps, but humans remain responsible for verification.

## Slide 5: Benchmark Lab

- Tests safe AI behavior across African election contexts, languages, and harm categories.
- Covers South Africa, Kenya, Nigeria, DRC, and Tanzania.
- Includes English, French, Swahili, isiXhosa, isiZulu, Afrikaans, and code-switched prompts.

**Speaker notes:**  
The Benchmark Lab is the research layer of the project. It asks whether AI systems behave safely outside English-only settings, especially when election risks appear in local languages and code-switched communication.

## Slide 6: Why It Matters / Next Steps

CivicGuard combines **Monitoring & Tooling** with **Evaluation & Benchmarking**.

Next steps:

- Supabase persistence
- Authenticated reviewer roles
- Partnerships with fact-checkers and civic organisations
- Real model evaluation APIs
- Expanded African-language benchmark coverage

GitHub: https://github.com/josuekb09/civicguard-africa  
Demo: https://civic-guard-africa.vercel.app

**Speaker notes:**  
The goal after the hackathon is to turn this prototype into a stronger civic safety workflow with real partners, persistent data, reviewer roles, and broader African-language evaluation. The current MVP shows the direction clearly and safely.