Revenue Rescue Agent
2026Autonomous AI agent
An agent that chases overdue invoices on its own. Gemini picks the offer, but plain code sets the discount limit, so it can never give away margin.
overview
Revenue Rescue Agent is an invoice recovery system. It ranks overdue invoices, predicts how likely each customer is to accept an offer, and negotiates payment over multichannel outreach.
the problem
Letting an LLM make financial decisions risks hallucinated terms and unsafe discounts. The system needed hard guardrails so the model could never issue terms nobody authorized.
how it works
Financial logic lives in deterministic verification services, separate from the prompt. The Gemini agent calls 5 specialized tools to score each debt, while strict mathematical floors protect margin and generate authenticated Razorpay payment links.
result
An offer-acceptance classifier trained on 30K synthetic invoice transactions drives dynamic discounts, turning delinquent accounts into completed Razorpay settlements.
- 30,000 synthetic transactions in the training set
- 5 autonomous agent tools
- Discounts can never breach the margin floor
built with
Python, FastAPI, Gemini API, PostgreSQL, Razorpay API, Docker