A modular Retrieval-Augmented Generation (RAG) system built from scratch to explore document indexing, semantic retrieval, prompt engineering, streaming responses, and production-ready AI architectures.

Overview

Mini RAG is a modular Retrieval-Augmented Generation (RAG) system built from scratch to explore how modern AI applications retrieve, reason over, and generate responses from private knowledge bases.

The project implements the complete RAG pipeline—from document ingestion and semantic retrieval to prompt engineering, conversation memory, and streaming LLM responses—while keeping every component modular and replaceable.

AI Pipeline

Documents
    ↓
Chunking
    ↓
Embeddings
    ↓
Vector Database
    ↓
Semantic Retrieval
    ↓
Prompt Builder
    ↓
LLM
    ↓
Streaming Response

Highlights

Retrieval-Augmented Generation (RAG)
Semantic Vector Search
Conversation-Aware Responses
Session Memory
Streaming LLM Output
Automatic Source Citation
Configurable Prompt Engineering
Native & LangChain Retrieval

Modular Architecture

The system is built around interchangeable providers and factories, making it easy to integrate new LLMs, embedding models, retrieval engines, vector stores, and document loaders with minimal changes to the rest of the codebase.

Future Work

Hybrid Search
Cross-Encoder Re-ranking
Ollama & OpenAI Providers
Metadata Filtering
Multi-Agent Integration
Tool Calling

Installation

github link: https://github.com/sayan-dey12/rag-system

See README.md file for more detailed description and Insatallation steps

Mini RAG