Projects

A two-stage training framework that first compresses a large "teacher" vision-language model into a smaller "student" using standard knowledge distillation, then further improves the student using reinforcement learning rewards.

This project presents VesselNet, a multi-pathway deep CNN for 3D hepatic vessel segmentation that classifies each voxel using three orthogonal 2D patches (sagittal/coronal/transverse) to better capture vessel structure in 3D.

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.

A GraphRAG system designed for Vision-Language Models (VLMs) to answer employee questions from policy/legal documents and visually rich PDFs. It combines vector retrieval over multimodal chunks with a knowledge graph to improve scope correctness, reduce wrong-but-similar retrieval, and produce explainable answers with citations and reasoning paths.

An n8n-based AI orchestration system for banking that routes each request to the most suitable path: direct response, ML workflow, text-only LLM, multimodal VLM, vector RAG, or GraphRAG. The system chooses the lowest-cost path that still gives enough accuracy and evidence, helping automate bank document work, policy Q&A, fraud review, credit risk analysis, and other internal workflows.