title	SummaPDF
emoji	📄
colorFrom	purple
colorTo	blue
sdk	docker
app_port	7860
pinned	false

PDF Summarizer

A web app that summarizes PDF documents using Hugging Face's BART-large-CNN model.

Tech stack

• FastAPI — async Python web framework
• HTMX — interactivity without writing JavaScript
• Tailwind CSS — utility-first styling (via CDN)
• Jinja2 — server-rendered templates
• pypdf — PDF text extraction
• transformers + PyTorch — running facebook/bart-large-cnn locally

Features

• Drag-and-drop PDF upload with file preview (name + size)
• Length presets (Brief / Balanced / Detailed) plus a manual min/max under "Advanced"
• Chunked summarization for long documents
• Result card with stats: source characters, summary words, compression %
• Copy-to-clipboard and .txt download
• Inline error reporting and a button-embedded loading state, all via HTMX
• Dark UI with gradient mesh background, Inter typography, subtle animations

Installation

git clone https://github.com/shexiee/summapdf.git
cd summapdf
pip install -r requirements.txt

The first run downloads the BART model (~1.6 GB) into the Hugging Face cache.

Running locally

uvicorn main:app --reload

Open http://localhost:8000.

Project layout

.
├── main.py                       # FastAPI app + summarization logic
├── templates/
│   ├── base.html                 # HTML shell (Tailwind + HTMX)
│   ├── index.html                # Upload form
│   └── partials/
│       ├── summary.html          # HTMX swap target on success
│       └── error.html            # HTMX swap target on failure
├── static/                       # Optional static assets
└── requirements.txt

How it works

1. The form POSTs a PDF to /summarize via HTMX (multipart/form-data).
2. pypdf extracts the document text.
3. The text is sliced into 1024-character chunks (BART's input limit).
4. Each chunk is summarized; the summaries are concatenated.
5. The server returns an HTML partial, which HTMX swaps into the page — no full reload.

Deployment

Any host that runs Python works (Render, Railway, Fly.io, a VPS). Example start command:

uvicorn main:app --host 0.0.0.0 --port $PORT

Note: the Tailwind CDN logs a console warning in production. For a polished deploy, swap it for a compiled stylesheet (npx tailwindcss -i input.css -o static/app.css --minify) and serve from /static.

Requirements

• Python 3.9+
• ~2 GB free disk for the model cache
• CPU works; a GPU dramatically speeds up summarization