Retrieval-Augmented Generation (RAG) System

My Experiments & Learning

A hands-on sandbox repository to explore and understand the fundamentals of Retrieval-Augmented Generation (RAG).

The goal of this repo is to experiment with document chunking, embeddings, vector search, and prompting LLMs with custom retrieved context.

---

What's Inside

• Document Ingestion & Chunking: Testing different chunk sizes and overlaps on sample text/PDFs.
• Embeddings & Vector Search: Generating embeddings and running similarity searches using a local vector store.
• Prompt Augmentation: Injecting retrieved context chunks into LLM prompts to see how response accuracy improves.

---

Tech Used

• Language: Python
• Orchestration: LangChain
• Document Processing: PyMuPDF
• Vector Store: ChromaDB
• LLM & Inference: Groq API
• Embeddings: HuggingFace (sentence-transformers)

---

🏗️ System Architecture

               ┌──────────────────────────┐
               │    Document Ingestion    │
               │ (PDF, Markdown, TXT, ...)│
               └────────────┬─────────────┘
                            │
                     [Text Splitter]
                            │
               ┌────────────▼─────────────┐
               │     Embedding Model      │
               └────────────┬─────────────┘
                            │
                     [Vector Embeddings]
                            │
               ┌────────────▼─────────────┐
               │      Vector Database     │
               └────────────┬─────────────┘
                            │
User Query ──► [Embed Query]│ (Top-K Similar Chunks)
                            ▼
              ┌───────────────────────────┐
              │    Prompt Construction    │
              │  (Query + Context Chunks) │
              └─────────────┬─────────────┘
                            │
              ┌─────────────▼─────────────┐
              │        LLM Engine         │
              └─────────────┬─────────────┘
                            │
                            ▼
                     Grounded Response

Setup & Try

1. Clone the repo

git clone [https://github.com/DevDreamer26/RAG.git](https://github.com/DevDreamer26/RAG.git)
cd RAG
---