Build a PDF Q&A Agent with RAG

Step 1: Install DependenciesStep 1: Install Dependencies

Step 1: Install Dependencies

Before writing any code, let’s install the required libraries.

pip install langchain openai chromadb pypdf langchain-community

Step 2: Load the PDF Document

PDFs aren’t plain text they contain structured data, formatting, and pages. So we need a tool to extract the actual readable text from each page.

LangChain provides a plugin called PyPDFLoader that turns PDFs into LangChain-friendly Document objects

from langchain_community.document_loaders import PyPDFLoader
loader = PyPDFLoader("Digital Transformation with Catalyst.pdf") #Load your document
documents = loader.load()

• "documents"is now a list of pages, each as a text chunk we can process.
• Stratus is Catalyst’s scalable object storage service for storing files like PDFs, images, or documents. It integrates seamlessly with serverless functions, enabling automated workflows and secure, cloud-based file management.
• Check out : Stratus Catalyst Native Object Storage

Step 3: Chunk the Text

Why chunk the document?

LLMs can’t handle huge amounts of text all at once. If we try to pass a 100-page document to GPT-4, it will either fail or truncate.

So we break the text into smaller, overlapping chunks just like scanning through parts of a book instead of reading the whole thing at once.

from langchain.text_splitter import RecursiveCharacterTextSplitter
splitter = RecursiveCharacterTextSplitter(chunk_size=500,
chunk_overlap=100, separators=["\n\n", "\n", ".", " "] ) # try to split at logical places
chunks = splitter.split_documents(documents)

Step 4: Convert Text Chunks into Embeddings

What are embeddings?

Think of embeddings as a way to represent text as numbers in a high-dimensional space

Similar pieces of text will have similar embeddings. This lets us search semantically, not just by keyword.

from langchain.embeddings import OpenAIEmbeddings
embedding_model = OpenAIEmbeddings(openai_api_key="your-openai-key")

Now we can convert any text (a chunk or a question) into a numerical vector for comparison.

Step 5: Store Chunks in a Vector Database (Chroma)

What is a vector store?

A vector store lets you store embeddings and search by similarity. When the user asks a question, we convert that into an embedding and search the database for chunks that are “close” to it in vector space.

We’ll use Chroma, a lightweight and fast vector database.

from langchain.vectorstores import Chroma
db = Chroma.from_documents(
    documents=chunks,
    embedding=embedding_model,
    persist_directory="vector_db"
)
db.persist()  # saves the data to disk

Step 6: Ask Questions Using

The heart of RAG

Now, we can build a function that:

• Embeds the user’s question
• Finds the most relevant chunks from the database
• Sends those chunks (as context) to the LLM
• Returns the answer           
   

from langchain.prompts import PromptTemplate
from langchain.chat_models import ChatOpenAI
def answer_question(question: str):
    # Step 1: Search the most relevant document chunks
results = db.similarity_search(question, k=4)
    context = "\n\n".join([doc.page_content for doc in results])
    # Step 2: Create a clear prompt with the context and question
prompt_template = """
    Use ONLY the information in the context to answer the question.
    Do not make anything up or include external knowledge.
    Context:
    {context}
    Question:
    {question}
    Answer:
    """
    prompt = PromptTemplate.from_template(prompt_template).format(
        context=context,
        question=question
    )
    # Step 3: Send prompt to LLM
model = ChatOpenAI(openai_api_key="your-openai-key")
    return model.predict(prompt)

Try it out!

response = answer_question("What risks were listed in the 2023 financial report?") print(response)

You’ll get a concise, factual answer pulled directly from the PDF content.

Response

This PDF is about **technological advancements, specifically focusing on chatbots**.

It discusses:

• The impact of technology (like smartphones) on daily life.
• The utility and functions of chatbots (e.g., consumer satisfaction, healthcare duties, data collection).
• The developer's perspective on building chatbots.
• The definition, capabilities, training, and embedding of chatbots within applications, mentioning "Catalyst" in this context.