reconFX

• 

  This is the main dashboard for the investigation results

Every month, finance teams close their books with numbers that don’t quite add up.

A variance appears. Someone investigates for a while, deadlines approach, and the difference gets parked in a clearing or suspense account. Month after month, these unexplained amounts accumulate until millions of dollars become nearly impossible to trace.

We built reconFX for the Autonomous CFO Challenge to change that.

reconFX acts like a forensic financial investigator. It starts by excavating historical unexplained balances, tracing transactions and evidence to identify recurring causes. It then uses that history to investigate the current month’s variance—testing hypotheses such as timing differences, approved exclusions, FX, duplicates, and misclassifications.

An LLM handles reasoning and evidence discovery, while a deterministic Python engine performs the actual financial calculations, preventing hallucinated numbers from reaching the ledger.

The result: explain what can be explained, recover what was missed, and escalate only what truly remains unresolved.

Built With

• ao
• javascript
• node.js
• python