FINANCE AND AUDIT

Find the invoices worth a second look.

Halcyon Industries is a synthetic company built for demos. Its accounts-payable team has thousands of invoices and a handful that matter. Three engines in one database narrow them down, and they agree.

STEP 1

A model scores every invoice

Train on past invoices using amounts, purchase-order match, how close the amount sits to an approval limit, self-approval, and the vendor’s category and country. The model reaches a ROC AUC of 0.96 on held-out invoices.

Plain-language health checks before you promote
Scores are written back to an ordinary table

osdbstore

Anomaly detection: model health and scored invoices1400 x 1000 pxDelete this block and drop in an Image or Video widget

osdbstore

Anomaly detection: graph path from vendor to employee1400 x 1000 pxDelete this block and drop in an Image or Video widget

STEP 2

The graph explains the link

A flagged vendor is only a number until you can see why. The relationship graph shows a path from the vendor to a bank account that an employee also holds.

Vendor, bank account, employee: the path is shown
Relationships come from the records, not from guesses

STEP 3

Documents add the context

Audit notes are searched by meaning. The answer names the same vendors and cites the passage it came from, so a reviewer can open the source.

Hybrid search over audit notes
Every answer shows its sources

osdbstore

Anomaly detection: cited answer from audit notes1400 x 1000 pxDelete this block and drop in an Image or Video widget

See it on your own invoices.

Start with the one-command stack and the sample data, then connect your own tables.