This project parses unstructured bank statement PDFs into structured financial data using YOLO-based layout detection, OCR, and LLM table extraction. It powers natural language querying via a RAG pipeline with Chroma/Faiss vector stores and autonomous agents built with AG2. The system categorizes income and expenses, generates monthly and yearly financial summaries, and provides a Streamlit web interface for end users. It supports both cloud LLMs (Gemini, OpenRouter) and local models via Ollama.