	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Google Gemini API > Document Intelligence Agent - Q and A over PDFs

Document Intelligence Agent - Q and A over PDFs

Author: Venkata Sudhakar

Enterprise teams spend hours reading through lengthy reports, contracts, and policy documents to find specific information. A document intelligence agent powered by Gemini can ingest an entire PDF, understand its structure, and answer precise questions with direct citations - functioning like a personal research assistant for any document.

In this tutorial, we build a document Q and A agent for ShopMax India. The agent ingests a business document (annual report or vendor contract) using Gemini Files API, then answers questions about it with page references and quoted evidence.

The below example shows the agent processing a vendor contract PDF and answering compliance questions with citations.

import os
import pathlib
from google import genai
from google.genai import types

client = genai.Client(api_key=os.environ["GEMINI_API_KEY"])

# Step 1: Upload the document using Files API
# For demo we use a locally saved PDF
pdf_path = pathlib.Path("shopmax_vendor_contract.pdf")

print("Uploading document to Gemini Files API...")
uploaded_file = client.files.upload(
    path=pdf_path,
    config=types.UploadFileConfig(
        mime_type="application/pdf",
        display_name="ShopMax Vendor Contract - Q1 2025"
    )
)
print(f"File uploaded: {uploaded_file.name}")
print(f"File URI: {uploaded_file.uri}")

# Step 2: Build the document reference part
doc_part = types.Part.from_uri(
    file_uri=uploaded_file.uri,
    mime_type="application/pdf"
)

# Step 3: Ask questions about the document
questions = [
    "What is the contract value and payment terms?",
    "What are the penalty clauses for delivery delays?",
    "What is the dispute resolution mechanism specified in this contract?"
]

system_prompt = """You are a contract analysis assistant for ShopMax India.
When answering questions:
1. Quote the relevant clause or section directly.
2. Mention the section/clause number if available.
3. If the information is not in the document, say so clearly.
4. Keep answers precise and legally accurate.
"""

for question in questions:
    response = client.models.generate_content(
        model="gemini-2.0-flash",
        contents=[
            types.Content(role="user", parts=[
                doc_part,
                types.Part(text=question)
            ])
        ],
        config=types.GenerateContentConfig(
            system_instruction=system_prompt
        )
    )
    print(f"Q: {question}")
    print(f"A: {response.text}")
    print()

It gives the following output,

Uploading document to Gemini Files API...
File uploaded: files/abc123xyz
File URI: https://generativelanguage.googleapis.com/v1beta/files/abc123xyz

Q: What is the contract value and payment terms?
A: Per Clause 4.1 (Financial Terms): The total contract value is Rs 48,00,000
(Forty-Eight Lakhs) for Q1 2025. Payment terms are net-30 from invoice date.
Clause 4.3 specifies: "Invoices must be submitted within 7 days of delivery
acceptance. Payment shall be made within 30 calendar days of invoice receipt."

Q: What are the penalty clauses for delivery delays?
A: Clause 8.2 (Liquidated Damages) states: "In the event of delivery delay
beyond the agreed schedule, the Vendor shall be liable for liquidated damages
at the rate of 0.5% of the delayed order value per day, subject to a maximum
cap of 10% of the total contract value."
For this contract, the maximum penalty is Rs 4,80,000.

Q: What is the dispute resolution mechanism?
A: Clause 14 (Dispute Resolution) specifies a three-tier process:
1. Amicable settlement within 30 days of written notice
2. Mediation through Mumbai Centre for International Arbitration (MCIA)
3. Arbitration under Indian Arbitration and Conciliation Act 1996
Jurisdiction: Mumbai High Court. Governing law: Laws of India.

Files API cleanup output,

File files/abc123xyz deleted from Files API

The Gemini Files API supports PDFs up to 2GB and retains files for 48 hours. For persistent document stores, re-upload files at session start or use Vertex AI RAG Engine for long-term indexed document retrieval. To handle multi-document questions (compare two contracts), pass multiple doc_part objects in the same contents list - Gemini can reason across all of them simultaneously.

Send your comments, suggestions or queries regarding this site to [email protected].