Avertix

Inspiration

The idea for NPA-Predict AI came from the rising problem of Non-Performing Assets (NPAs) in Indian banks. Traditional systems react after defaults occur, causing huge financial losses. I wanted to build an AI system that predicts NPAs months in advance, helping banks act early and avoid defaults.

What I Learned

While building the project, I learned how financial systems work, how to engineer features for time-series data, and how to apply machine learning for default prediction. I also learned the importance of explainability and user experience in fintech tools.

How I Built It

I generated a 36-month synthetic loan dataset using Python (pandas and numpy) with both static and dynamic features. Then I performed feature engineering to calculate rolling averages, late payment counts, and volatility. I trained models like Logistic Regression, Random Forest, and XGBoost to predict loan default risk. Finally, I built a Streamlit dashboard that shows each loan’s NPA Risk Score, allows sorting, and simulates scenarios such as missed payments to update risk in real time.

Challenges Faced

The main challenges were the lack of real banking data, model explainability, and ensuring realistic behavior in the synthetic dataset. I also had to overcome platform limits for code submission by hosting my prototype externally.

Prototype

• GitHub: github.com/mihirphalke1/avertix
• Live Demo: avertix.vercel.app
• Pitch Video: YouTube

Conclusion

NPA-Predict AI helps banks move from reactive to proactive risk management by combining AI, financial data, and interactive visualization. This project taught me how technology can solve real problems in the banking sector and the importance of building interpretable, impactful AI systems.