Tessera DB
PX-01
Solution
Follow the money over one copy of the data.
A fraud or AML investigation needs the network of accounts, the rows behind every transfer, and the history of what changed and who changed it. Usually that is a graph database, a warehouse and an audit system that disagree with each other. Here it is one store.
01
04
What the work needs.
01
The network and the rows at once
Circular flows and shared attributes are graph questions; balances and amounts are SQL questions. Both are asked of the same records, so the answers agree.
02
What was known, and when
An alert is judged on what the data said when it fired. Valid time and transaction time on every record let you replay it as of that moment.
03
Who touched the evidence
Every write records who made it and through which door, and the audit trail is hash-chained, so an edited entry is named when the chain is checked.
04
Exact identity lookups
Which customers share this phone number is an index lookup, not a scan. Every candidate is read back and compared, so a hash collision never returns a wrong row.
02
04
What Tessera DB adds on Enterprise+.
01
Banking pack
A BIAN service-domain catalogue, FIBO entity validation, liquidity (LCR, NSFR) and SAR report models. The fraud graph algorithms, such as smurfing and money-mule bridging, are on Enterprise.
02
Sanctions screening
A name-matching engine with a small built-in sample list. Loading the official OFAC, EU, UN and HMT lists is planned, not shipped.
03
Regulatory report drafts
Drafts for GDPR, Basel / BCBS 239 and the EU AI Act, built from the data you hold. They are drafts for your compliance team, not filings.
04
People
A dedicated solutions architect, on-premise installation including air-gapped sites, training and onboarding.
03
04
Our own fraud benchmark, and what it does not prove.
Workload | Per alert (p50) | Whole book | Recall / precision |
|---|---|---|---|
F1 · circular flow | 1.7 ms | 8.2 s | 1.000 / 0.995 |
F2 · structuring | 0.013 ms | 459 ms | 1.000 / 1.000 |
F3 · ultimate beneficial owner | 0.001 ms | 8.9 ms | 1.000 / 1.000 |
F4 · shared-attribute identity | 0.014 ms | 27.1 ms | 1.000 / 1.000 |
← Swipe for the rest of the table →
← Drag for the rest of the table →
Self-reported, with no auditor. A generated bank book of 497,469 records (21,398 customers, 32,094 accounts, 4,994 companies, 406,288 transfers) on an Apple M2 Pro, in the engine and single-threaded, not over HTTP. Timings vary by up to 15% between runs. The same author wrote the planted fraud and the detector, so the scores show that the engine runs these rules, not that the rules catch criminals.