UttaranHazarika/africa-agent

# AFRICA – Autonomous Financial Risk & Credit Decision Agent AFRICA is a multi-agent AI system that evaluates loan applications by combining classical machine learning, policy-based rules, vector memory retrieval, and LLM reasoning. The goal is to demonstrate how an **agentic AI system** can make structured, explainable credit decisions while maintaining governance and auditability. ## Overview The system processes loan applications through a sequence of agents orchestrated with **LangGraph**. Each agent performs a specific task and passes structured outputs to the next stage. Workflow: User Request → FastAPI API → LangGraph Workflow → Risk Agent → Policy Agent → Memory Agent → Decision Agent → Governance Logging → API Response ## Agents Risk Scoring Agent Computes a credit risk probability using a classical machine learning model (Logistic Regression). Policy & Compliance Agent Applies rule-based checks such as credit score thresholds, loan-to-income ratios, and minimum age requirements. Memory Agent Retrieves similar historical credit cases from a **Chroma vector database** to provide contextual decision support. Decision Reasoning Agent Uses an LLM to synthesize risk scores, policy outcomes, and retrieved cases to produce a structured decision and explanation. Governance Agent Logs decision traces and maintains audit-ready records. ## License This project is intended for research and educational use.