← Back to projects
ML and AI Chatbot for Financial Domain
A Gradio-based app for RAG banking chatbot using LangChain + Llama 2 + FAISS. This project is a banking-focused data science that explores fraud detection and risk analytics across multiple datasets and modeling approaches for banking chatbot.
Data AnalysisFinancialBankingMLAI Chatbot
Problem
Detecting fraudulent banking transactions is hard because real-world datasets are often highly imbalanced and require heavy preprocessing/feature work. The overall goal is to support fraud detection/risk assessment with practical modeling and analysis workflows
Approach
- Multi-model fraud scoring + UI: The 'Credit Card Fraud Detection' app supports multiple models and provides 'risk assessment & explanation' with plots like feature-importance bars and correlation heatmaps.
- Dataset-driven projects: The project lists several datasets and methods, including credit-card fraud, a separate (credit-risk/feature engineering) workflow, and IEEE-CIS with LightGBM-style modeling and engineered features.
- Extra module (chatbot): A banking Q&A chatbot using RAG (retrieval + generation) with LangChain, Llama 2, and FAISS
Results
LightGBM training reports a best validation AUC of about 0.9666 and an out-of-fold ROC AUC calculation of 0.9665. A Gradio-based app for RAG banking chatbot using LangChain + Llama 2 + FAISS.
Highlights
- Evidence the models can separate fraud vs non-fraud pretty well.
- Real-world fraud problems, not just toy examples.
- A banking Q&A chatbot (RAG)