Skip to content

lab

RevLedger

Tool-grounded answers over a deterministic subscription revenue engine

The agent refuses the first question because no data is loaded, then the sample goes in and the same question is answered from the computed schedule rather than from the model.

What it does

Prepaid class packages bring cash in before the service is delivered. RevLedger reads sales and attendance CSVs, then computes journal entries, a deferred-revenue roll-forward, breakage and a P&L summary. A React interface exposes the reports and CSV or JSON downloads.

The accounting engine uses Decimal amounts and eight exact validation checks, including global and per-transaction balance, customer conservation and over-consumption. These checks establish internal consistency under the encoded assumptions; they are not independent certification of accounting-standard compliance.

A LangChain agent has six focused tools over the computed results. Python calculates the figures; the model selects tools and explains their output. The chat records the tools used and supports conversation memory. Processing and reports work without an API key; live chat requires one.

Stack

  • Python
  • FastAPI
  • pandas
  • LangChain
  • React
  • Docker

The repository tests synthetic fixtures and provided sample CSVs. The sample's provenance is not documented as synthetic.