crypt.ml
A risk score with a reasoning trail.
A multi-agent AML prototype connecting transaction rules, ML scoring, graph analysis, and an operations dashboard.
Live dashboard demo
What I built
A flag is more useful when an analyst can see which signals produced it.
- Designed five cooperating agents around an asynchronous event bus.
- Used XGBoost and SHAP contributions for inspectable risk scoring, alongside graph-based pattern detection.
- Built an analyst dashboard and SAR drafting flow. Displayed statistics are demo content, not independently measured production results.
Tools: Python, FastAPI, React, XGBoost, SHAP, NetworkX, Docker.
Website & project links
- crypt.ml website ↗https://crypt-ml-rho.vercel.app
- Public repository ↗https://github.com/https-shubhamsahu/crypt.ml
About the builder
Shubham Sahu is an engineering student and builder in Mumbai, working across embedded systems, robotics, AI tools, and web and mobile applications. He co-founded Aurigen.
Shubham Sahu’s official profile and social accounts ↗