kpatc/microlend-recommender

Financial product recommendation engine for African SMEs — collaborative filtering, matrix factorization, neural CF, hybrid model with cold-start support, MLflow model registry, and FastAPI web interface. # MicroLend Recommender > Financial product recommendation engine for African SMEs — collaborative filtering, matrix factorization, neural CF, and hybrid approaches with cold-start support and MLflow model registry. --- ## Table of Contents - Business Problem - Architecture - Dataset - Models - Results - Quick Start - CLI Reference - MLflow Tracking & Model Registry - REST API - Tech Stack --- ## Business Problem Over **50 million SMEs** across Sub-Saharan and North Africa lack access to tailored financial products. Microfinance institutions (MFIs) typically offer the same 2–3 products to all clients — personalization is near zero and adoption rates stay chronically low. **This project applies recommendation system techniques to microfinance**, matching each SME to the financial products most likely to fit their profile and needs: | Without Recommender | With Recommender | |---|---| | Same products pushed to everyone | Personalized ranking per SME | | Loan officers rely on intuition | Data-driven similarity matching | | Cold-start = no offer | Cold-start bootstrapped from profile | | High default risk from product mismatch | Risk-adjusted scoring | **Expected impact:** 25–40% improvement in product adoption rates, 15–20% reduction in defaults through better product-client matching. --- ## Architecture ``` Raw Data (CRM log + SME profiles) │ ▼ ┌─────────────────────────────┐ │ Data Layer │ │ sme_profiles.csv │ │ sme_financial_profile.csv │ ──► DataLoader ──► Merged SME features │ product_interactions.csv │ ──► build_ratings_long() ──► User-item matrix │ product_catalog.csv │ └─────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────┐ │ Model Layer │ │ │ │ ┌─────────────┐ ┌──────────────┐ ┌───────────┐ │ │ │ User-based │ │ Item-based │ │ Matrix │ │ │ │ CF │ …