Get Answers From Your PDF — Azure OpenAI and Langchain

In this article, I’ll walk you through all the steps required to query your PDFs and get response out of it using Azure OpenAI.

Let’s get started by importing the required packages.

Import Required Packages

Here are the packages which we need to import to get started:

from dotenv import load_dotenv
from langchain.document_loaders import UnstructuredFileLoader
from langchain.text_splitter import CharacterTextSplitter
from langchain.embeddings import OpenAIEmbeddings
from langchain.vectorstores import Chroma
from langchain.chains import RetrievalQA
from langchain.llms import AzureOpenAI

Set Environment Variables

First of all, we need to set few variables with information from Azure portal and Azure OpenAI Studio:

OPENAI_API_TYPE = "Azure"
OPENAI_API_VERSION = "2022-12-01"
OPENAI_API_BASE = "ENDPOINT"
OPENAI_API_KEY = "API_KEY" 
DEPLOYMENT_NAME = "DEPLOYMENT_NAME_FROM_AI_STUDIO"

If you are not sure how to grab above values, I would recommend you watch my below video on this.

Next, we will go ahead and use above variables to set environment variables:

from dotenv import load_dotenv

os.environ["OPENAI_API_TYPE"] = OPENAI_API_TYPE
os.environ["OPENAI_API_VERSION"] = OPENAI_API_VERSION
os.environ["OPENAI_API_BASE"] = OPENAI_API_BASE
os.environ["OPENAI_API_KEY"] = OPENAI_API_KEY
load_dotenv()

Load PDF

Next, we will load our PDF using UnstructuredFielLoader class which comes with Langchain.

loader = UnstructuredFileLoader('Sample.pdf')
documents = loader.load()

Split Documents Into Chunks

Once the PDF is loaded, next we need to divide our huge text into chunks. You can define chunk size based on your need, here I’m taking chunk size as 800 and chunk overlap as 0.

text_splitter = CharacterTextSplitter(chunk_size=800, chunk_overlap=0)
texts = text_splitter.split_documents(documents)

Prepare Model And Embeddings

Till here, we are ready with our data. Now the only thing remaining is, generating embedding, associating them with text, select a large language model and stuff the data into it. All these steps can be done in just few lines of code as shown below:

embeddings = OpenAIEmbeddings()
doc_search = Chroma.from_documents(texts,embeddings)
chain = RetrievalQA.from_chain_type(llm=AzureOpenAI(model_kwargs={'engine':'text-davinci-002'}),chain_type='stuff', retriever = doc_search.as_retriever())

Create Query And Get Response

Now we are ready to ask questions and get response.

query = 'What are the effects of homelessness?'
chain.run(query)

On executing above query, you will get the response based on the content mentioned in your document :)

I hope you find this walkthrough useful.

If you find anything, which is not clear, I recommend you watch my video recording, which demonstrates this flow from end-to-end.

Above flow is for Azure OpenAI but if you would like to query your PDF using OpenAI then I would recommend you read my another article from here.