Featured Project

RAG-Based Financial Chatbot

Retrieval-Augmented Generation pipeline for contextual Q&A over financial documents using vector embeddings, OCR, and LLMs.

OpenAISentence TransformersFAISSCLIPPaddleOCRScikit-learnPython

View on GitHub

Detailed Overview

A production-grade RAG pipeline that enables intelligent Q&A over financial reports. The system processes unstructured documents using OCR, embeds them with sentence transformers, stores them in a FAISS vector index, and uses LLMs to generate source-grounded responses with financial ratio analysis and trend identification.

Problem Statement

Financial analysts spend hours manually reading lengthy PDF reports to extract insights. There was no automated way to query financial documents with contextual understanding.

Technical Solution

Built a RAG pipeline that ingests financial PDFs via PaddleOCR, chunks and embeds the content using Sentence Transformers, indexes it in FAISS, and uses OpenAI to generate accurate, source-grounded responses with citations.

Architecture & Data Flow

Document ingestion → OCR (PaddleOCR) → Chunking → Embedding (Sentence Transformers) → Vector storage (FAISS) → Query embedding → Similarity search → LLM (OpenAI) → Response with sources