Boyang Yan

PDF summarizer application

3 min read

Objective

PDF summarizer application that downloads, extracts, and summarizes research papers(https://arxiv.org/) locally using Gemma3 (4b) + LangChain + Ollama.

How it works ?

PDF Summarization application is designed to process research papers efficiently using Gemma 3, split them into manageable chunks, and generate structured summaries.

Key Components of this PDF Summarizer

  • User Uploads PDF — The user provides an ArXiv PDF URL in Streamlit, which sends a request to the FastAPI backend.
  • Text Extraction — The backend downloads the PDF and extracts text using PyMuPDF (Fitz).
  • Chunking with LangChain — RecursiveCharacterTextSplitter divides the text into manageable chunks for processing
  • Summarization with Gemma 3 — Each chunk is sent to Ollama, where Gemma 3 generates summaries in parallel for efficiency.
  • Final Output in Streamlit — Summaries are merged into a structured format using Pydantic , displayed in the UI, and available for download.

Implementation Guide

Let’s go through a step-by-step breakdown of this PDF extract and Summarization application.

Step 1. Create a new directory for your project

mkdir pdfSummarizerApp
cd pdfSummarizerApp

Step 2: Initialize the project using uv (this creates pyproject.toml and .venv)

uv init

Step 3: Add dependencies to pyproject.toml

uv add fastapi uvicorn requests langchain pydantic pymupdf streamlit ollama httpx
  • FastAPI: For building the backend API.
  • Uvicorn: An ASGI server to run FastAPI applications.
  • Requests: For handling HTTP requests.
  • LangChain: For managing text processing and interaction with the language model.
  • Pydantic: For data validation within FastAPI.
  • PyMuPDF: For extracting text from PDFs.
  • Streamlit: For creating the frontend user interface.
  • Ollama: For running the Gemma 3 model locally.​[using 27B for demo]
  • httpx: For making asynchronous HTTP requests.

Step 4: Activate the virtual environment (Linux/macOS)

source .venv/bin/activate

Step 5: Run a quick test

python -c "import fastapi, uvicorn, requests, langchain, pydantic, fitz, streamlit, httpx; print('All good!')"

Step 6: Setup ollama and Download Gemma3

yanboyang713@Meta-Scientific-Linux ~/projects/pdfSummarizerApp (git)-[master] % ollama list
NAME         ID              SIZE      MODIFIED
gemma3:4b    a2af6cc3eb7f    3.3 GB    8 hours ago
yanboyang713@Meta-Scientific-Linux ~/projects/pdfSummarizerApp (git)-[master] % ollama show gemma3:4b
  Model
    architecture        gemma3
    parameters          4.3B
    context length      131072
    embedding length    2560
    quantization        Q4_K_M
 
  Capabilities
    completion
    vision
 
  Parameters
    stop           "<end_of_turn>"
    temperature    1
    top_k          64
    top_p          0.95
 
  License
    Gemma Terms of Use
    Last modified: February 21, 2024

import packages

import os
import logging
import requests
import fitz
import asyncio
import json
import httpx
from concurrent.futures import ThreadPoolExecutor
from fastapi import FastAPI
from pydantic import BaseModel
from langchain.text_splitter import RecursiveCharacterTextSplitter
import ollama

handle long text when doing extraction

Reference List

  1. https://medium.com/google-cloud/running-googles-gemma-3-llm-langchain-locally-with-ollama-with-full-code-a57c94754393
  2. https://github.com/arjunprabhulal/gemma3_pdf_summarizer

Graph View