Inspiration Modern digital payment processing demands fast, reliable fraud prevention without creating friction for legitimate cardholders. Our team wanted to build an MVP beta prototype to explore how modern machine learning models and lightweight web APIs can collaborate to score financial transactions instantly and provide clear visual feedback to risk analysts.
What it does FinGuard AI is a functional MVP beta prototype for real-time payment fraud detection:
Ingests card transaction features and calculates immediate fraud risk probabilities.
Implements a 3-tier dynamic decision routing mechanism:
Low Risk: Approved.
Medium Risk: 3D-Secure challenge / secondary verification required.
High Risk: Immediate payment block.
Features a simple, interactive Streamlit audit dashboard allowing risk officers to manually test transactions, view telemetry, and review model outputs.
How we built it
Model Training: Trained a classification model using Kaggle credit card transaction data, handling extreme class imbalance.
Backend Scoring: Built an API using FastAPI, Uvicorn, and Pydantic schema validation to run low-latency inference.
Dashboard Interface: Developed an interactive front-end application with Streamlit.
Source Code: https://github.com/umar-git-cmd/FinGuardAI/tree/master
Challenges we ran into
Handling severe class imbalance during model training, as fraudulent operations account for less than 0.2% of total transactions.
Connecting the local FastAPI backend with the Streamlit dashboard and testing end-to-end data flow within a tight hackathon timeframe.
Accomplishments that we're proud of
Successfully created and connected a working end-to-end beta prototype within 24 hours.
Verified that transaction risk scores are returned instantly with clear status indicators.
Kept the codebase modular, documented, and easy to run locally via requirements.txt.
What we learned
How to evaluate fraud models using Precision-Recall curves rather than misleading raw accuracy metrics.
Rapid prototyping best practices connecting FastAPI backend services with Streamlit frontends.
Key operational concepts behind FinTech risk engines and payment routing.
What's next for FinGuardAI
Refining feature engineering and testing more advanced classification algorithms.
Deploying the backend and frontend to a cloud environment.
Integrating database persistence to store transaction audit histories.
Built With
- ai
- fastapi
- joblib
- kaggle
- numpy
- pandas
- pydantic
- python
- scikit-learn
- streamlit
- uvicorn
- web
Log in or sign up for Devpost to join the conversation.