
Fraud needs detection engineers
An agent writes and tests AML and fraud detections on your data. You own every rule it writes.
Points are accounts. Lines join each to its transactions and sessions across the operator’s sources. Teal marks a reversed transaction or a proxy login. The rule flags accounts where both pile up.
On one remittance operator’s data it read 301,099 accounts and flagged five. Four were fraudulent.
The run took five hours over 100 iterations and used no labelled cases. The full run
4 of 5
Flagged accounts confirmed fraudulent on one operator's data. One known case missed.
5 hours
From export to a scored rule. No implementation project, no procurement cycle.
0 labels
Rules are written from raw exports. Confirmed cases only sharpen the scoring.
Yours
Every rule is a file your team reads, replays and owns. Nothing lives in our database.
From $5,000
A year. The same volume was quoted elsewhere at $42,000 to $107,573.
Drafted
A draft report per alert, formatted for FinCEN, goAML or the NCA. Your officer files.
No dashboards to learn. Install atrosa, point it at an export, and write, test and run detection rules right from the terminal. Plain commands you can script and automate.
# Run the agent on an export: up to 100 iterations, scored without labels $ atrosa hunt --data-dir ./exports --max-iterations 100 --production # Replay the kept rules over the export as a stream, without acting $ atrosa sentinel --mode simulate --dry-run # Watch a directory for new batches and alert on hits $ atrosa sentinel --mode watch --watch-dir ./incoming # List the drafted reports waiting for your compliance officer $ atrosa sar list --status draft $
One model per customer, trained on your data inside your environment and used to write rules.
Send one export and you will have results within 24 hours of our receiving it. ATROSA is in private beta.