AfriGen is a synthetic data infrastructure platform for African ML engineers, designed to generate, govern, and validate high-fidelity datasets grounded in real-world statistics.

Note: First load may take 30–60 seconds (Render free tier spins down when inactive). For full OpenMetadata catalog integration, run locally with your own instance.

African machine learning engineers face a critical data gap. There is almost no quality training data that reflects African contexts. Our health systems, demographics, languages, and realities. Models trained on Western data consistently underperform when deployed in Africa.

AfriGen fixes this.

AfriGen transforms real-world African statistics into validated, production-ready synthetic datasets for machine learning.

It uses World Health Organization (WHO) Global Health Observatory and World Bank data to ground Gemini AI generation in real statistical distributions, ensuring outputs reflect actual African health and demographic realities instead of random synthetic patterns.

Each dataset is then scored for statistical fidelity, automatically registered in OpenMetadata with full lineage and schema tracking, and made discoverable through a live catalog.

Before any dataset reaches a training pipeline, it passes through a data quality layer that checks completeness, detects anomalies, scans for sensitive data, and evaluates model readiness with AI-assisted recommendations.

WHO API + World Bank API

↓

Ground Truth Statistics

↓

Gemini AI (generates synthetic rows matching real distributions)

↓

Fidelity Scoring Engine

↓

OpenMetadata Catalog (schema + lineage + governance)

↓

AfriGen UI (generate, validate, download)

Every generated dataset is scored 0–100 using a two-layer algorithm:

Layer 1 — Gemini AI Evaluation (80% of score)

Gemini evaluates 50 randomly sampled rows against 4 criteria:

• Feature relationship coherence - do the columns make sense together? (40 pts)
• Realistic distributions - are values varied, not uniform or random? (30 pts)
• Risk factor consistency - do risk indicators align logically? (20 pts)
• Logical coherence - are there contradictions across rows? (10 pts)

Layer 2 - Completeness Check (20% of score)

A statistical completeness function scans every cell in the full dataset:

function getCompleteness(csv: string) {
  const rows = csv.split('\n').filter(Boolean)
  let total = 0
  let filled = 0
  rows.forEach(r => {
    r.split(',').forEach(cell => {
      total++
      if (cell.trim() !== '') filled++
    })
  })
  return (filled / total) * 100
}

Final Score Formula: finalScore = (0.8 × GeminiScore) + (0.2 × completenessScore)

This penalizes missing values even when AI-evaluated rows look good.

Before Gemini generates a single row, AfriGen fetches real statistics for the requested country and domain from:

• WHO Global Health Observatory - disease prevalence, mortality rates, health indicators
• World Bank Open Data - population statistics, demographic distributions, economic indicators

These real statistics become ground truth constraints passed directly to Gemini as part of the generation prompt. This is what separates AfriGen from a random data generator, every synthetic row is anchored to real African data.

AfriGen uses OpenMetadata as its core governance layer via the REST API.

Every generated dataset is automatically:

• Registered as a Table Entity under the afrigen-synthetic Database Service
• Given full column schema with data types and descriptions
• Tagged with Tier3 governance tag
• Stored with a complete description including domain, country, fidelity score, and the exact prompt used to generate it
• Tracked with lineage - the Gemini AI Generator pipeline is registered as the source, with an edge connecting it to every output table
• Discoverable in the AfriGen Catalog which pulls live from OpenMetadata API

Database Service: afrigen-synthetic (CustomDatabase) Database: default Schema: synthetic_datasets - healthcare_dataset__nigeria (Table) - malaria_dataset__kano (Table) - ...

Pipeline Service: afrigen-pipelines (CustomPipeline) Pipeline: gemini-generator Outputs: - healthcare_dataset__nigeria (Lineage Edge) - malaria_dataset__kano (Lineage Edge)

• Node.js 18+
• Docker Desktop (for OpenMetadata)
• Gemini API Key - get one free here

git clone https://github.com/[your-username]/afrigen.git
cd afrigen

npm install

Create a .env.local file in the root of the project:

GEMINI_API_KEY=your_gemini_api_key_here
OPENMETADATA_URL=http://localhost:8585
OPENMETADATA_TOKEN=your_openmetadata_bot_token_here

cd afrigen-data-platform
docker compose -f docker-compose.yml up --detach

Wait 7-8 minutes for all containers to start, then open: http://localhost:8585 - login with admin@open-metadata.org / admin

Get your bot token:

• Go to Settings → Bots → ingestion-bot → copy the JWT token
• Paste it as OPENMETADATA_TOKEN in your .env.local

# Start the app first
npm run dev

# Then visit this URL once to create the service, database and schema
http://localhost:3000/api/setup-om

http://localhost:3000

Generate your first African dataset. Watch it appear in OpenMetadata automatically.

• Not just a data generator - It is a full data infrastructure platform
• Real statistics from WHO and World Bank ground every dataset
• Automatic governance - every dataset is cataloged, tagged, and lineage-tracked
• Fidelity scoring tells you exactly how trustworthy your data is before training
• Built specifically for African ML engineers who have been underserved by existing tools.

MIT - built with ❤️ for Africa