~/projects/recovery-agent.md
AI Payment Recovery Agent
An AI agent prototype that classifies failed payments, selects a recovery strategy, simulates outcomes, and adapts later decisions using historical performance.
- 6payment-failure types modelled
- ≤3attempts per transaction
- JSONdiagnosis, strategy, confidence
Repository (opens in a new tab)All projects
How it works
- Failed payment
- Past attempts + stats
- LLM diagnosis
- Pick strategy
- Simulate outcome
- Record result ↺ learn: back to “Past attempts + stats”
What it does
- Uses LLM reasoning to classify payment-failure context and choose a recovery action.
- Records simulated outcomes and feeds strategy-performance history into subsequent decisions.
- Combines a web dashboard with a FastAPI backend, SQLite, and a Groq-hosted LLM.