🚀 Featured Projects

---

🏦 Finance RAG Chatbot

Description: A Retrieval-Augmented Generation (RAG) chatbot designed to retrieve and summarize information from complex financial terms and conditions (PDF).

• Key Features:
  • Semantic Chunking: Implemented custom text splitting based on financial clauses (Articles/Chapters) to maintain document structure using regex.
  • Advanced Reranking: Applied MMR (Maximal Marginal Relevance) and LLM-based scoring (Upstage Solar Pro 2) to enhance retrieval accuracy and reduce redundancy.
  • User Interface: Developed an interactive web interface using Streamlit for seamless PDF uploads and real-time Q&A.
• Tech Stack: Python 3.12, LangChain, Upstage Solar API, ChromaDB, Streamlit
• Link: View on GitHub

---

📊 Explainable Credit Default Prediction (XAI)

Description: A comparative study on credit card default prediction using SHAP and DiCE to achieve both model interpretability and actionable counterfactual insights.

• Key Features:
  • Dual XAI Approach: Leveraged SHAP for feature attribution (Logistic Regression) and DiCE for counterfactual explanations (XGBoost).
  • Model Comparison: Analyzed linear vs. non-linear decision boundaries to evaluate how different models interpret financial risk.
  • Counterfactual Analysis: Generated “what-if” scenarios to identify minimal feature changes required to flip a prediction from ‘Default’ to ‘Non-default’.
• Tech Stack: Python, Scikit-learn, XGBoost, SHAP, DiCE, Pandas
• Link: View on GitHub