RAG Implementation using LangChain

This project demonstrates how to build a basic Retrieval-Augmented Generation (RAG) pipeline leveraging LangChain, Google Generative AI's Gemini-2.0 model, and ChromaDB. The pipeline loads, processes, embeds, and indexes PDF documents for retrieval-augmented question answering.

Table of Contents Overview

Features

Tech Stack

Setup Instructions

Workflow Details

Example Usage

Notes

Overview The notebook implements a pipeline where a user’s question is answered using information retrieved from a PDF document. It loads a PDF from a URL, splits the content into manageable chunks, creates vector embeddings, stores them in a vectorstore (Chroma), and then uses Gemini-2.0 via LangChain for retrieval-augmented QA.

Features PDF loading from URL

Splitting documents into overlapping text chunks

Creating embeddings using Google Generative AI

Vector storage and semantic search with ChromaDB

Retrieval-Augmented QA using Gemini-2.0 (LangChain integration)

Example Q&A workflow

Tech Stack Python (Colab/Jupyter)

LangChain

langchain-google-genai

ChromaDB

pypdf

Google Generative AI (Gemini-2.0/Flash Model)

Setup Instructions In a Colab/Jupyter notebook:

Install Required Packages

python !pip install -q --upgrade google-generativeai langchain-google-genai !pip install langchain_community !pip install pypdf !pip install chromadb API Key Setup Ensure your Google Generative AI API Key is stored in Colab’s userdata as GOOGLE_API_KEY:

python from google.colab import userdata API_KEY = userdata.get('GOOGLE_API_KEY') Workflow Details Model Initialization Create a Gemini-2.0 Flash chat model using your API key.

python from langchain_google_genai import ChatGoogleGenerativeAI model = ChatGoogleGenerativeAI(model="gemini-2.0-flash", api_key=API_KEY) Document Loading Load a PDF document from the web.

python from langchain_community.document_loaders import PyPDFLoader pdfloader = PyPDFLoader('<PDF_URL>') loaded_pdf_doc = pdfloader.load() Splitting Text Break the loaded document into text chunks for embedding.

python from langchain.text_splitter import RecursiveCharacterTextSplitter recursive_splitter = RecursiveCharacterTextSplitter(chunk_size=300, chunk_overlap=50, separators=['\n\n', '\n', ' ']) chunks = recursive_splitter.split_documents(loaded_pdf_doc) Embedding Creation Generate embeddings for each chunk using Google Generative AI.

python from langchain_google_genai import GoogleGenerativeAIEmbeddings embeddings = GoogleGenerativeAIEmbeddings(model='models/embedding-001', google_api_key=API_KEY) Vector Storage Store the embeddings using Chroma as a vector database.

python from langchain.vectorstores import Chroma vector_store = Chroma.from_documents(documents=chunks, embedding=embeddings) Retriever Configuration Build a retriever interface for similarity search.

python vector_index = vector_store.as_retriever(search_kwargs={'k':8}) Setup RetrievalQA Chain Compose a RAG pipeline using LangChain’s RetrievalQA.

python from langchain.chains import RetrievalQA rag = RetrievalQA.from_chain_type( model, retriever = vector_index, return_source_documents=True ) Ask Questions Run questions through the RAG system.

python question = 'Can you talk about what is there in property1?' response = rag({"query": question}) print(response) Example Usage python

question = 'Can you talk about what is there in property1?' response = rag({"query": question}) print(response) This will output the answer along with the supporting source documents from the indexed PDF.

Notes Replace <PDF_URL> with the actual URL of your target PDF.

Make sure your API key quota is sufficient for your usage.

To load local PDFs, use the appropriate local file path with PyPDFLoader.

You can customize the chunk_size, overlap, retriever parameters, or the Gemini model version.

Chroma runs locally by default in Colab/Jupyter. For larger-scale or persistent storage, refer to Chroma documentation for alternative setups.

End of README