01 / TEAM / HACKATHON PROJECT
Sentinel AML.
AI that follows the evidence.
An investigation workspace that turns a question into an analytical plan, explainable risk, and a trace you can inspect.
01 / THE CONTEXT
The problem worth solving.
Transaction analysis needs more than an anomaly score. An investigator needs to know which data was examined, why a result was flagged, and how to reproduce the finding.
The approach
A query-driven workspace chooses relevant analytical tools, combines rules and model signals, and keeps source evidence connected to each explanation and review action.
02 / UNDER THE HOOD
How the pieces connect.
03 / ENGINEERING JUDGMENT
The decisions behind the build.
Plan around the question
Select aggregation, exploration, feature, model or graph tools based on intent rather than running every tool for every query.
Ground the explanation
Tie generated findings to detector signals and source data so an analyst can inspect the evidence.
Keep humans in the loop
Separate risk prioritization from a final regulatory decision. The workspace supports investigation and review.
04 / WHAT TO INSPECT
Evidence, and its boundaries.
- 5.1M+ records across documented analytical workspaces
- Source-backed traces and review surfaces
- An AWS deployment variant documented in a companion repository
Dataset scale is documented project evidence, not a customer or production-traffic claim. The AWS variant is the same product, not a second project.
Inspect the repository AWS deployment variant05 / THE TAKEAWAY
What I’m taking forward.
Good AI tooling makes uncertainty and provenance visible. An explanation becomes useful when someone can verify its source.
KEEP EXPLORING