📄 PDF to Social Media Post Generator

📝 Overview

This tool allows users to upload a PDF (i.e. research paper, article, or report) and automatically generates a Social Media post that summarizes the content. The app uses meta-llama/llama-3.3-8b-instruct:free via OpenRouter to produce compelling, professional summaries, while securely logging usage events to a Google Cloud bucket. It works seamlessly both locally and on Hugging Face Spaces.

✨ Features

📄 PDF Text Extraction: Extracts text from uploaded PDFs using PyPDF2
🚨 PII Checker: Ensures not PII is sent to the LLM
🤖 AI-Powered Summarization: Uses LLM models for post generation
✅ Relevance Assurance: Keeps key concepts from the original document
🎯 Format Optimization: Produces clean, ready-to-publish SocialMedia text
🌐 Cloud-Based Logging: Stores usage logs in a GCS bucket (Vertex AI compatible)
🧑‍💻 Works Locally and in Hugging Face Spaces

🚀 Getting Started

✅ Prerequisites

Python 3.8 or higher
OpenRouter API key → https://openrouter.ai
(Optional for local use) GCP Service Account JSON key for logging

🔧 Installation

# 1. Clone the repository
git clone https://huggingface.co/spaces/your-username/post_generator
cd post_generator

# 2. Create a virtual environment with uv package 
- Install UV if not already installed
  curl -LsSf https://astral.sh/uv/install.sh | sh

- Create a virtual environment with UV
  uv venv .venv

- Activate the virtual environment
  source .venv/bin/activate

- Install required dependencies
  uv pip install openai gradio python-dotenv PyPDF2 presidio-analyzer spacy
  
  python -m spacy download en_core_web_lg

🔐 API and Credential Setup

📌 1. Create a `.env` file:

OPENROUTER_API_KEY=your-openrouter-key
GOOGLE_APPLICATION_CREDENTIALS=/absolute/path/to/your/gcp_key.json

On Hugging Face Spaces, you don’t need the .env file — add secrets via Settings > Secrets.

📌 2. Hugging Face Secret Configuration:

Key	Value
`OPENROUTER_API_KEY`	OpenRouter API key
`GOOGLE_APPLICATION_CREDENTIALS`	Paste raw JSON from GCP key

▶️ Running the App

Locally:

python app/main.py

Your app will be available at: http://127.0.0.1:7860

Hugging Face Spaces:

Just push the code — it auto-deploys!

💡 How It Works

Upload PDF: User selects a file
Text Extraction: PDF parsed using PyPDF2
Post Generation: Text sent to LLM via OpenRouter
Cloud Logging: Logs are pushed to your GCS bucket (if credentials exist)
Display Result: The Social Media post appears, ready to copy

📁 Project Structure

pdf-to-socialmedia-app/
├── app/
│   ├── main.py              # Gradio UI + routing
│   ├── llm_integration.py   # LLM logic via OpenRouter
│   ├── pdf_processing.py    # PDF extraction
│   ├── export_handler.py    # Text export logic
│   └── analytics.py         # Cloud/local logging abstraction
    └── explore.ipynb        # Jupyter Notebook for quick exploration
├── .env                     # (local only) API keys
├── requirements.txt         # Python dependencies
├── README.md  
├── LICENSE                  # It contains the terms of the MIT License
├── analytics_logs/          # Local logs (if GCS is not used)

⚙️ Configuration Options

Change the LLM Model

Edit llm_integration.py:

model = "meta-llama/llama-3.3-8b-instruct:free"

Explore more models at: https://openrouter.ai/models

Customize Generation Style

Edit the instruction prompt in llm_integration.py to change tone, length, or format of the post.

🔒 Security Notes

The app never stores the PDFs uploaded beyond session memory.
OpenRouter keys and GCP credentials are stored in .env (local) or Hugging Face Secrets (hosted).
GCS logs are append-only and used solely for usage analytics.

🛠️ Future Features

🔐 Authentication (e.g., for private use or teams)
🔄 Direct Social Media Posting (API integration)
📑 Support for DOCX, PPTX files
💾 Enhanced cloud storage of generated posts

📄 License

MIT License — see LICENSE file for details.

🤝 Contributing

All PRs, suggestions, and feedback are welcome!
Open an issue or fork and submit a pull request.