TraceWeave — Multi-Source AML Detection
A source-agnostic anti-money-laundering triage engine. It reads messy bank, call-detail and social exports, infers what each column actually means, links the same person across all three sources, and ranks them by risk. Seven unsupervised model families — time correlation, OddBall network analysis, Isolation Forest, Benford's law, structuring, behavioural and chain detection — are fused by Borda count and gated by a sufficiency engine that never runs a detector the data cannot honestly support. Validated blind on the SAML-D benchmark and shipped as a PySide6 desktop app with SHA-256 chain-of-custody sealing.
Python
scikit-learn
RapidFuzz
Pandas
IsolationForest
Graph Analysis
PySide6
Pytest