🦜 LangChain Practice — GenAI + RAG Pipeline

A hands-on end-to-end GenAI project built with LangChain, Groq (LLaMA 3.3), FAISS VectorStore, and Streamlit — featuring a fully deployed Question Answering bot with LangSmith monitoring.

---

🚀 Live Demo

---

📌 Project Overview

This project demonstrates a complete Retrieval-Augmented Generation (RAG) pipeline using LangChain, where a PDF document is ingested, embedded, stored in a vector database, and queried using a powerful LLM — all wrapped in a clean Streamlit UI.

The bot acts as an AI Engineer assistant, answering questions strictly based on the provided PDF context.

---

🏗️ Architecture

PDF Document
     │
     ▼
[Data Loading] ──► PyPDFLoader / TextLoader
     │
     ▼
[Text Splitting] ──► CharacterTextSplitter / RecursiveCharacterTextSplitter
     │
     ▼
[Embedding] ──► HuggingFace Embeddings (sentence-transformers)
     │
     ▼
[VectorStore] ──► FAISS Index (faiss_idx/)
     │
     ▼
[Retriever] ──► Similarity Search
     │
     ▼
[LLM] ──► Groq API → LLaMA 3.3 (70B)
     │
     ▼
[Response] ──► Streamlit UI (app.py)
     │
     ▼
[Monitoring] ──► LangSmith Tracing

---

🧩 Project Structure

LangChain_practice/
│
├── LangChain/
│   ├── Data_Embedding/          # Embedding + VectorStore notebooks
│   ├── faiss_idx/               # Persisted FAISS vector index
│   │
│   ├── Data_extraction.ipynb    # RAG pipeline: load → split → retrieve
│   ├── llmModel.ipynb           # RAG + Groq LLaMA 3.3 integration
│   ├── genai_App.ipynb          # Full GenAI app prototype (notebook)
│   │
│   ├── app.py                   # 🚀 Streamlit app — deployed QA Bot
│   ├── document.pdf             # Source PDF for knowledge base
│   ├── Synopsis_Major_project.pdf
│   ├── test.txt                 # Test ingestion file
│
├── Requirements.txt
├── .gitignore
└── README.md

---

⚙️ Tech Stack

Layer	Tool / Library
Framework	LangChain
LLM	Groq API — LLaMA 3.3 (70B Versatile)
Embeddings	HuggingFace sentence-transformers
Vector DB	FAISS (persisted locally)
Frontend	Streamlit
Monitoring	LangSmith
Language	Python 3.10+

---

🔄 Pipeline Breakdown

1️⃣ Data Ingestion & Splitting

• Loaded PDF using PyPDFLoader
• Split into chunks using RecursiveCharacterTextSplitter
• Explored chunk size, overlap, and splitting strategies

2️⃣ Embedding & VectorStore

• Generated embeddings using HuggingFaceEmbeddings
• Stored and persisted vectors using FAISS (faiss_idx/)
• Reloaded index for retrieval without re-embedding

3️⃣ RAG + LLM (llmModel.ipynb)

• Built a RetrievalQA chain using LangChain
• Connected FAISS retriever with Groq's LLaMA 3.3
• Extracted context-aware answers from the PDF

4️⃣ Streamlit App (app.py)

• Built a conversational QA interface
• System prompt: "Act as an AI Engineer to answer questions based on the given context"
• Used ChatGroq with streaming support
• Deployed live on Streamlit Cloud

5️⃣ Monitoring with LangSmith

• Integrated LangSmith tracing for all chain runs
• Tracked inputs, outputs, token usage, and latency

---

🛠️ Setup & Installation

1. Clone the repo

git clone https://github.com/ravicoder01/LangChain_practice.git
cd LangChain_practice

2. Install dependencies

pip install -r Requirements.txt

3. Set environment variables

Create a .env file in the root:

GROQ_API_KEY=your_groq_api_key
LANGCHAIN_API_KEY=your_langsmith_api_key
LANGCHAIN_TRACING_V2=true
LANGCHAIN_PROJECT=langchain-practice

4. Run the Streamlit app

cd LangChain
streamlit run app.py

---

📦 Requirements

langchain
langchain-community
langchain-groq
faiss-cpu
sentence-transformers
pypdf
streamlit
python-dotenv
langsmith

---

💡 Key Learnings

• How RAG solves LLM hallucination by grounding responses in real documents
• FAISS index persistence — embed once, query many times
• Difference between CharacterTextSplitter and RecursiveCharacterTextSplitter
• Using Groq for blazing-fast LLaMA inference
• LangSmith for production-grade observability

---

👨‍💻 Author

Ravi Roy B.Tech CSE (AI & ML) | IILM University, Greater Noida

---

"RAG is not just a technique — it's the bridge between static LLMs and dynamic real-world knowledge."