Custom RAG Document Chat System Recently,

Custom RAG Document Chat System

Recently, I built a Custom Retrieval-Augmented Generation (RAG) system that allows users to upload their own documents and interact with them through natural-language questions.

The goal was to understand how modern AI document-chat systems work internally — from document processing and chunking to embeddings, vector search, and LLM-based responses.

What I Built

📄 Document Upload — Users can upload PDF and DOCX documents.

🔍 Text Extraction — Extracts readable text from uploaded documents.

✂️ Document Chunking — Splits large documents into smaller overlapping chunks for better retrieval.

🧠 Embeddings — Converts document chunks into vector representations.

🗄️ Vector Storage — Uses Qdrant to store and search document embeddings.

🔎 Semantic Search — Retrieves the most relevant document chunks based on the user's question.

🤖 LLM Integration — Uses retrieved context to generate grounded answers.

💬 Document-Based Chat — Users can ask questions and receive answers based on their uploaded documents.

RAG Pipeline

Upload → Extract → Chunk → Embed → Store → Retrieve → Generate

This project helped me understand the complete RAG pipeline rather than treating RAG as just an API call to an LLM.

Tech Stack

Backend: ASP.NET Core Web API, C# Document Processing: PDF & DOCX text extraction Embeddings: Ollama + nomic-embed-text Vector Database: Qdrant AI: Retrieval-Augmented Generation (RAG), LLM API Testing: Swagger Frontend: React.js