Skip to content
ML Academy

S24 · adr simulation

Fuel invoice reconciliation, agent or workflow

People, data and outputs in this scenario are simulated. Each stage is recorded when you commit; the reveal comes after.

  • 4stages
  • 40 minabout
  • 0stages committed

The brief

The synthetic airline's finance team reconciles monthly fuel invoices from suppliers at 14 stations against delivery records (station, date, litres, price per litre, currency). Discrepancies above a tolerance set by finance must be flagged with a short explanation for the supplier manager, who decides what to send.

Two prototypes were built on 200 synthetic invoices, 12 of which use layouts never seen before. Tools are mocks; no real supplier is contacted. All results are simulated.

  • Workflow W: parse known layouts with templates; match lines to deliveries in SQL with tolerances; convert currency from the daily rate table; flag by rule; an LLM writes the explanation text only, from the structured mismatch.
  • Agent A: an LLM with tools read_invoice_pdf, query_deliveries, get_fx_rate, draft_supplier_note decides the steps itself.

Stage 1 of 4

Before comparing prototypes, decide which parts of the task need judgment at all.

Which steps should be deterministic code rather than model judgment? Select all that apply.

Sign in to play scenarios (answers are recorded before the reveal).