A compliance engine engineered for how crime actually moves.
Vigilic is an end-to-end ML platform for AML compliance. It ingests every transaction, scores it across 275 behavioral features spanning 14 risk categories, and generates ranked, explainable, regulator-ready investigation cases — not alerts.
Four layers. One pipeline from raw transaction to regulator-ready case.
Every transaction. Every channel. Every context.
Core banking feeds, ACH, wire, card, mobile, and online channels flow into a unified event stream. Customer reference data, counterparty records, and external enrichment sources (sanctions, PEPs, adverse media, geographic risk, corporate registries) are joined in real time.
Behavioral features across 14 risk categories, computed per transaction.
The feature engine is the heart of Vigilic. Every transaction is transformed into a dense behavioral vector spanning magnitude, velocity, temporal patterning, counterparty graphs, geographic risk, channel behavior, structuring signatures, network topology, unsupervised anomalies, typology-specific fingerprints, external enrichment, composite ensembles, and adverse media correlations.
Category-specific models, unified into one decision.
Each risk category has its own specialized model — supervised, semi-supervised, and unsupervised where appropriate. An ensemble layer reconciles their outputs into a single risk judgment with confidence bounds and full feature attribution, so every case carries its own explainability trail.
Not alerts. Cases.
What reaches your BSA team is a complete investigation package: a ranked case with typology attribution, supporting network graph, counterparty context, adverse media links, prior-activity timeline, and suggested SAR narrative. Everything an investigator — or an examiner — needs to see the full picture.
The 14 risk categories that drive every Vigilic decision.
Rule-based transaction monitoring typically relies on 15 to 30 hand-crafted rules. Vigilic's engine computes 275 behavioral features — organized into 14 risk categories — on every transaction. The categories below are not a feature list; they are risk lenses, each driven by its own domain-specialized model.
Full 275-feature technical briefing available under mutual NDA
The complete engineering-grade walkthrough — feature-level specifications, model governance, validation methodology, and deployment architecture — is shared with qualified institutional prospects, investors, and partners under NDA.
Cross-institution collaboration, secure by design.
Sophisticated laundering operations deliberately fragment activity across institutions — precisely because no single bank sees the whole picture. That blind spot is structural in every monitoring system on the market today.
Vigilic Phase 2 closes it. Our cryptographic protocol lets participating institutions collectively detect coordinated networks — without exposing customer PII, without centralizing sensitive data, and without regulatory exposure for any participant.
The result: typologies that have been invisible to community banks for decades — smurfing rings operating across 20 institutions, trade-based networks, funnel accounts — become visible for the first time.
Built to examiner standards, from day one.
Vigilic is architected from the ground up for financial-institution deployment. Our engineering discipline reflects the expectations of bank examiners, internal audit teams, and model risk management functions.
Model governance
Every model in Vigilic is documented with its training data lineage, feature dependencies, validation metrics, drift monitoring, and explainability methodology — aligned to SR 11-7 model risk management guidance. Every case output carries its own feature attribution trail.
Data protection
Institution data never leaves institution-controlled boundaries without explicit cryptographic protection. Our Phase 2 cross-institution protocol is designed so that no participant — including Vigilic — has visibility into another participant's raw transaction or customer data.
Regulator alignment
Our design-partner relationship and founding team's 20+ years of AML operations experience at major U.S. institutions shape how we build. Every feature, case output, and audit artifact is designed to survive an examiner review.
Responsible AI
We don't write a single line of production code — or train a single model — on real bank customer data in AI development environments. Development uses synthetic data; training uses institution-controlled pipelines.
See the engine in motion.
A 30-minute walkthrough with a founder. No sales pitch — a working demo of case generation, feature attribution, and the Phase 2 protocol.