How to Build a PDF Chatbot with Langchain

To build a PDF chat using Langchain, load the PDF with PyPDFLoader, split the text into chunks, create embeddings with OpenAIEmbeddings, and use Chroma as a vector store. Then, connect a RetrievalQA chain with an OpenAI language model to answer questions based on the PDF content.

Syntax

This is the basic syntax to build a PDF chat with Langchain:

• PyPDFLoader: Loads PDF files and extracts text.
• CharacterTextSplitter: Splits text into smaller chunks for processing.
• OpenAIEmbeddings: Converts text chunks into vector embeddings.
• Chroma: Stores and searches embeddings efficiently.
• ChatOpenAI: The language model that generates answers.
• RetrievalQA: Combines retrieval and question answering.

python

from langchain.document_loaders import PyPDFLoader
from langchain.text_splitter import CharacterTextSplitter
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.vectorstores import Chroma
from langchain.chat_models import ChatOpenAI
from langchain.chains import RetrievalQA

# Load PDF
loader = PyPDFLoader('your_file.pdf')
docs = loader.load()

# Split text
text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
documents = text_splitter.split_documents(docs)

# Create embeddings
embeddings = OpenAIEmbeddings()
vectorstore = Chroma.from_documents(documents, embeddings)

# Setup language model
llm = ChatOpenAI(temperature=0)

# Create retrieval QA chain
qa = RetrievalQA.from_chain_type(llm=llm, retriever=vectorstore.as_retriever())

Example

This example shows how to build a simple PDF chatbot that answers questions from a PDF file named sample.pdf. It loads the PDF, splits the text, creates embeddings, and runs a chat query.

python

from langchain.document_loaders import PyPDFLoader
from langchain.text_splitter import CharacterTextSplitter
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.vectorstores import Chroma
from langchain.chat_models import ChatOpenAI
from langchain.chains import RetrievalQA

# Load PDF file
loader = PyPDFLoader('sample.pdf')
docs = loader.load()

# Split text into chunks
text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
documents = text_splitter.split_documents(docs)

# Create embeddings and vector store
embeddings = OpenAIEmbeddings()
vectorstore = Chroma.from_documents(documents, embeddings)

# Initialize chat model
llm = ChatOpenAI(temperature=0)

# Create retrieval-based QA chain
qa = RetrievalQA.from_chain_type(llm=llm, retriever=vectorstore.as_retriever())

# Ask a question
query = "What is the main topic of the document?"
answer = qa.run(query)
print("Answer:", answer)

Output

Answer: The main topic of the document is ... (depends on PDF content)

Common Pitfalls

1. Not splitting text properly: Large PDFs need chunking to avoid exceeding token limits.

2. Missing API keys: Ensure your OpenAI API key is set in environment variables.

3. Using incompatible vector stores: Use Chroma or other supported vector stores for embeddings.

4. Forgetting to install dependencies: Install langchain, chromadb, and pypdf.

python

## Wrong: Not splitting text (may cause errors or poor results)
loader = PyPDFLoader('sample.pdf')
docs = loader.load()
# Using docs directly without splitting

## Right: Split text into chunks
text_splitter = CharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
documents = text_splitter.split_documents(docs)

Quick Reference

• Load PDF: PyPDFLoader('file.pdf').load()
• Split Text: CharacterTextSplitter(chunk_size=1000, chunk_overlap=100)
• Embeddings: OpenAIEmbeddings()
• Vector Store: Chroma.from_documents(docs, embeddings)
• QA Chain: RetrievalQA.from_chain_type(llm, retriever)

Key Takeaways

Load and extract text from PDFs using PyPDFLoader before processing.

Split large text into chunks to fit language model token limits.

Use OpenAIEmbeddings and Chroma vector store to create searchable document vectors.

Combine a chat model with a retriever in RetrievalQA for question answering.

Always set your OpenAI API key and install required packages before running.