AI / LLM
Billing Agent
Natural-language billing analysis
- Role
- Founder & Senior Software Engineer
- Period
- 2024 – Present
Overview
I built an LLM-powered billing agent so school staff can ask questions about invoices, balances, and payment history in plain language. The agent uses application tools to retrieve live billing data, then summarizes aging, outstanding balances, and recent activity. It sits on top of the existing ledger rather than replacing it — retrieve, analyze, and explain, with the database remaining the source of truth.
Outcomes
- Natural-language interface over production billing data
- Tool-using agent that retrieves invoices, balances, and payment history
- Designed to analyze and explain without writing a second source of truth
- Built against the same multi-tenant billing model used by Toddly schools
Architecture
Interface
- Natural-language chat
- Staff workspace
Agent
- LLM
- Tool calling
- Retrieval
Tools
- Invoices
- Balances
- Payment history
Source of truth
- MySQL billing data
- Stripe events
A chat UI sends a question to an agent loop. The model calls application tools that query billing tables, then returns a grounded answer with the retrieved context.
Tech stack
- Python
- LLMs
- Tool calling
- Flask
- MySQL
- Stripe