Cricket-RAG-Search

A beginner RAG (Retrieval-Augmented Generation) project built with LangChain — ask a question in the terminal, and it finds relevant passages from a cricket document using semantic search, then uses an LLM to generate a direct answer grounded in that context.

How it works

1. Load & split — a .txt document is loaded and chunked (TextLoader + RecursiveCharacterTextSplitter)
2. Embed & store — chunks are embedded (sentence-transformers/all-MiniLM-L6-v2) and stored in FAISS
3. Retrieve — user's query is embedded and matched against stored chunks via similarity search
4. Generate — retrieved chunks + query are passed to an LLM (Llama-3.1-8B-Instruct via Hugging Face) to produce a natural-language answer

Example

Enter your query (or type 'exit' to quit): Who is known as the God of Cricket?

Answer: Sachin Tendulkar is known as the "God of Cricket."

Tech stack

LangChain · Hugging Face (embeddings + hosted LLM) · FAISS · Python

Setup

git clone <your-repo-url>
cd <repo-folder>
python -m venv venv
venv\Scripts\activate        # Windows
pip install langchain langchain-community langchain-huggingface langchain-text-splitters faiss-cpu sentence-transformers torch python-dotenv

Add a .env file with your Hugging Face token:

HUGGINGFACEHUB_API_TOKEN=hf_your_token_here

Run:

python EmbeddedModels/document_similarity.py

What I learned

Document chunking, embeddings, vector similarity search, and the RAG pattern — grounding an LLM's answers in retrieved context instead of its own training data.

---

Built as a hands-on learning project while studying LangChain and RAG fundamentals.