aktech-io/AthenaCreditScore

AI-powered credit scoring platform for African SMEs and individuals — LightGBM + rule scorecard + LLM overlay, Kenyan market (TransUnion + Metropol CRBs) # NemoScore — Credit Intelligence for Nemo Neobank > Formerly the Athena Credit Initiative. Internal wire-level identifiers (Go module paths, > `athena_db`, `athena-net`, `athena.exchange`, service DNS names) intentionally keep the > legacy name — they are shared contracts with the deployed LMS. **AI-powered credit scoring for African SMEs and individuals — built on LightGBM, local-first LLM overlay, and MLflow.** > ⚠️ Parts of this README predate the Go rewrite and NemoScore phases — `CLAUDE.md` and > `docs/nemoscore-audit.md` are the authoritative, current references. --- ## Architecture ``` Kong Gateway (:80) ├── /api/auth → athena-java-service:8080 ├── /api/v1/crb → athena-java-service:8080 (champion-challenger → python) ├── /api/v1/credit-* → athena-python-service:8001 └── /api/v1/credit-reports → athena-python-service:8001 (API-key secured) athena-python-service:8001 — FastAPI + MCP, scoring engine, MLflow integration athena-java-service:8080 — Spring Boot, CRB/Mifos clients, auth, RabbitMQ pub postgres:5432 — 23-table schema (core + MLOps + feature store + portal) rabbitmq:5672 — messaging for scoring triggers and notifications mlflow:5000 — model registry and experiment tracking prometheus:9090 + grafana:3000 — metrics and alerting ``` --- ## Quick Start ```bash # 1. Configure environment cp .env.example .env # Edit .env — set JWT_SECRET, OPENAI_API_KEY (or LLM_PROVIDER=local), DB passwords # 2. Start all services docker-compose up -d # 3. Initialise the database (schema applied automatically via Docker entrypoint) docker-compose logs postgres | grep "database system is ready" # 4. Verify health curl localhost # Kong gateway curl localhost # Python service curl localhost # Java service curl localhost # MLflow # 5. R …