LangPilot：基于 LangChain 与 ChromaDB 的 RAG PDF 问答系统

RAG-Powered PDF Question Answering with LangChain & ChromaDB

A Retrieval-Augmented Generation (RAG) pipeline that turns any PDF into an intelligent Q&A system — powered by LangChain, ChromaDB vector embeddings, and Groq LLM inference.

How It Works · Features · Architecture · Quick Start · Tech Stack

🎯 Why LangPilot?

Extracting specific answers from large PDF documents is tedious and time-consuming. LangPilot automates this with a production-grade RAG pipeline:

1. Extract text from any PDF document
2. Chunk it into semantically meaningful segments
3. Embed chunks as vectors and store in ChromaDB
4. Retrieve relevant context at query time
5. Generate grounded answers via LLM — with citations from the source document

The result: a conversational chat interface where you can ask natural language questions and get accurate, context-aware answers.

✨ Features

🏗️ Architecture

┌─────────────────┐
│   PDF Document   │
└────────┬────────┘
         │
         ▼
┌─────────────────────────────────────────────────┐
│  1. EXTRACT  ─  extract.py                      │
│     PyMuPDF (fitz) → Extracted_Text.txt          │
└────────────────────────┬────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────┐
│  2. CHUNK  ─  chunks.py                          │
│     RecursiveCharacterTextSplitter → Chunks.pkl  │
│     (size: 500 | overlap: 50)                    │
└────────────────────────┬────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────┐
│  3. EMBED  ─  embeddings.py                      │
│     HuggingFace all-mpnet-base-v2                │
│     → ChromaDB (chroma_langchain_db/)            │
└────────────────────────┬────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────┐
│  4. RETRIEVE + GENERATE  ─  app.py               │
│     ChromaDB Retriever → Groq LLM               │
│     (llama-3.1-8b-instant, temp=0.0)             │
│                    ↕                             │
│           Streamlit Chat UI                      │
└─────────────────────────────────────────────────┘

🚀 Quick Start

# Clone the repository
git clone https://github.com/prasenjitshinde11/langpilot.git
cd langpilot

# Create virtual environment
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Set your Groq API key
export GROQ_API_KEY="your-groq-api-key"

# Step 1: Extract text from PDF
python extract.py

# Step 2: Chunk the extracted text
python chunks.py

# Step 3: Generate embeddings & store in ChromaDB
python embeddings.py

# Step 4: Launch the chat interface
streamlit run app.py

🛠️ Tech Stack

📁 Project Structure

langpilot/
├── app.py              # Streamlit chat UI + RAG query pipeline
├── extract.py          # PDF → plain text extraction
├── chunks.py           # Text → overlapping chunks
├── embeddings.py       # Chunks → vector embeddings → ChromaDB
├── chroma_langchain_db/  # Persistent ChromaDB storage
├── requirements.txt
└── README.md

🔑 Key Engineering Decisions

• RAG over Fine-Tuning — Retrieval-Augmented Generation provides accurate, source-grounded answers without the cost and complexity of model fine-tuning
• Local Embeddings — all-mpnet-base-v2 runs on CPU, eliminating API costs for the embedding step
• Groq for Inference — Ultra-fast LLM inference with llama-3.1-8b-instant at zero temperature for deterministic, factual responses
• Persistent ChromaDB — Embeddings are stored on disk, so you only need to process the PDF once
• Overlapping Chunks — 50-character overlap preserves context across chunk boundaries for better retrieval accuracy

🧪 Concepts Demonstrated

This project demonstrates practical knowledge in:

• 🔹 Retrieval-Augmented Generation (RAG) pipeline design
• 🔹 Vector embeddings and similarity search
• 🔹 LangChain orchestration (chains, retrievers, prompt templates)
• 🔹 ChromaDB vector database management
• 🔹 LLM integration via API (Groq)
• 🔹 Document processing and text chunking strategies

📬 Contact

Prasenjit Shinde — CS Undergrad @ SIRT Bhopal

If you found this project useful, consider giving it a ⭐