FraudShield

Inspiration

Financial fraud is rising rapidly, and many people—even those with experience managing their finances—don’t realize how easily unusual transactions can slip by unnoticed. A single wrong click or overlooked charge can lead to significant money loss.

We wanted to build something practical, fast, and useful for everyday people: a tool that examines a person’s transaction history and immediately alerts them to suspicious activity.

FraudShield was created to help users understand fraud, detect it early, and stay protected.

What it does

FraudShield allows users to: Upload a CSV file containing their transaction history Automatically analyze every transaction using custom rule‑based anomaly detection Flag risky transactions with a risk score Use Gemini AI to generate human‑readable explanations of why each transaction might be fraudulent View a summary report that highlights: Total transactions Number of suspicious items Risk distribution Important behavioral patterns

How we built it

FraudShield was built using a modern full‑stack architecture combining a Next.js + TypeScript + Tailwind CSS frontend with a FastAPI + Python backend. The frontend provides a clean drag‑and‑drop CSV upload interface, responsive dashboard, and integrated API calls for uploading and analyzing transactions. The backend handles CSV ingestion, validation, anomaly detection, and integrates with the Gemini API to generate natural‑language fraud explanations. All transaction data, analysis results, and reports are stored in an SQLite database, creating a complete pipeline from upload → analysis → explanation → results.

Challenges we ran into

Throughout development, we faced issues such as CORS errors between frontend and backend running on different LAN IPs, messy and inconsistent CSV formats from users, and the difficulty of designing a rule‑based anomaly system that didn’t over‑flag transactions. Integrating Gemini explanations into the analysis pipeline introduced asynchronous complexity, while syncing the frontend UI with backend processing required careful planning. We also had to ensure our database structure remained clean, avoided duplicates, and handled multiple uploads smoothly.

Accomplishments that we're proud of

Our team built a fully functional end‑to‑end fraud detection system featuring real‑time AI explanations, a clean and intuitive upload interface, and robust CSV validation. We achieved seamless integration across Python, Next.js, and Gemini, producing explainable fraud detection rather than black‑box outputs. The system reliably identifies suspicious behavior, categorizes anomalies, and presents them in a user‑friendly dashboard , all built collaboratively within the hackathon.

What we learned

We learned how to design AI prompts for real‑time financial analysis, handle unpredictable user‑generated CSV data, and resolve security/CORS issues that can break uploads.

Built With

• fastapi
• gemini
• next.js
• python
• tailwind
• typescript