Gblack98/msme-alternative-credit-scoring

Alternative credit scoring for MSMEs in Africa using mobile money & behavioral data — Supabase + FastAPI + ML # 🌍 MSME Alternative Credit Scoring — Africa > Real-time credit scoring API for micro/small businesses in Africa using **alternative data** (mobile money, behavioral signals) — because **80% of MSMEs are unbanked** and excluded from traditional credit. ## Problem Traditional credit scoring requires: bank statements, tax records, collateral, credit history. In Sub-Saharan Africa, **60-80% of adults lack formal banking** — but they do have: - 📱 Mobile money history (Wave, M-Pesa, Orange Money, MoMo) - 📞 Telecom behavioral data - 🛒 E-commerce transaction history - 📍 Geolocation patterns ## API — Score a Business in Real-Time ```bash curl -X POST api.your-domain.com -H "Content-Type: application/json" -d '{ "business_id": "BIZ-001", "country_code": "SN", "sector": "retail", "years_in_operation": 2.5, "monthly_revenue_usd": 800, "requested_amount_usd": 1500, "loan_purpose": "inventory", "mobile_money": { "total_transactions_90d": 45, "total_volume_usd_90d": 3200, "has_regular_income": true } }' ``` **Response:** ```json { "credit_score": 680, "risk_band": "B", "decision": "approved", "max_loan_amount_usd": 1500, "recommended_rate_pct": 12.0, "explanation": [ "✅ High mobile money activity — strong financial engagement", "✅ Business established for 2.5 years", "✅ Regular income pattern detected" ] } ``` ## Stack ``` FastAPI → REST API (scoring endpoint, <100ms latency) Supabase → PostgreSQL + RLS + real-time audit log LightGBM → ML scoring model (trained on historical repayment data) dbt → Feature engineering from raw transactions ``` ## Countries Supported Senegal 🇸🇳 · Côte d'Ivoire 🇨🇮 · Nigeria 🇳🇬 · Kenya 🇰🇪 · Ghana 🇬🇭 · Cameroon 🇨🇲 ## Setup ```bash pip install -r requirements.txt cp .env.example .env # Add SUPABASE_URL, SUPABASE_KEY supabase db push # Apply migrations uvicorn app.main:app --reload ``` ## Author **Ibrahima Gabar Diop** — GitHub · Kaggle