f/Sentinel

GitHub Repo README
Live Demo

Inspiration

Online marketplaces are growing faster than their fraud defenses and cybersecurity. I wanted to build a lightweight yet intelligent system that flags risky transactions before trust is broken.

What it does

This project analyzes transaction data and returns a fraud risk score, a clear decision (Safe/Review/Block), and explainable reasons behind the flag (velocity, amount anomaly, behavior shift).

How we built it

Trained a fraud detection model using real transaction data
Exposed predictions via a FastAPI backend
Built a clean frontend demo to simulate marketplace transactions
Added rule-based and machine learning explanations for transparency

Challenges we ran into

Severe class imbalance (fraud is rare)
Avoiding “black box” predictions
Designing a demo that feels real but stays simple

Accomplishments that we're proud of

Real-time fraud scoring API
Explainable predictions suitable for trust & compliance
A deployable, end-to-end system within hackathon scope

What we learned

I learned that fraud detection is as much about interpretability as accuracy - simple models and strong signals outperform complex black boxes. I also noticed that UX matters even in backend-heavy security tools.