Document Q&A

Working with Hugging Face

Jacob H. Marquez

Lead Data Engineer

What is document question and answering?

Answers questions from document content
Requires a document and a question
Provides direct or paraphrased answers

Question: "What is the total revenue of Q3?"

A document

Use cases for document Q&A

Legal, finance and support use cases

📑 Legal: Identify contract clauses

💰 Finance: Extract key figures

🤓 Support: Retrieve answers from manuals

Automating HR queries with document Q&A

📄 Info stored in US-Employee_Policy.pdf

🤖 Build a system to extract answers

🕑 Save HR time and effort

HR team is overwhelmed

Extracting text with pypdf

from pypdf import PdfReader


# Load the PDF file
reader = PdfReader("US-Employee_Policy.pdf")


# Extract text from all pages
document_text = ""
for page in reader.pages:

    document_text += page.extract_text()

Welcome to the US Employee Policy document...

Creating a Q&A pipeline

# Load the question-answering pipeline
qa_pipeline = pipeline(
    task="question-answering",
    model="distilbert-base-cased-distilled-squad")


question = "How many volunteer days are offered annually?"


# Get the answer from the QA pipeline
result = qa_pipeline(question=question, context=document_text)

print(f"Answer: {result['answer']}")

Answer: 1

Bringing it all together

📄 Use PdfReader from pypdf to load and read PDF files
🔎 Extract text with .pages and .extract_text() into document_text
🤔 Set up a question-answering pipeline
❓ Pass a question and contextto the pipeline
⏰ Wrap into functions to automate queries

Document Q&A

HR team building company culture

Let's practice!

Working with Hugging Face