📰 AI News Headline Generator

An end-to-end Natural Language Processing (NLP) web application that generates concise headlines from long-form news articles. Built from scratch using a Sequence-to-Sequence (Seq2Seq) architecture with Bahdanau Attention, trained on a dataset of 300,000+ news samples.

🛑 Live Demo https://huggingface.co/spaces/Ritee17/Headlie_Generator

🚀 Features

• Abstractive Summarization: Generates new sentences rather than just extracting existing ones.
• Attention Mechanism: Uses Bahdanau (Additive) Attention to focus on relevant parts of the source text during generation.
• Custom Inference Engine: Implements a recursive decoder loop (Greedy Search) for autoregressive text generation.
• Web Interface: Clean, responsive UI built with Flask and HTML/CSS for real-time interaction.

🛠️ Tech Stack

• Deep Learning: TensorFlow, Keras (LSTM, GRU, Custom Layers).
• Backend: Python, Flask.
• Frontend: HTML5, CSS3.
• Data Processing: Pandas, NumPy, NLTK.
• Hardware Optimization: Mixed Precision Training (FP16) for RTX 3050 GPU.

📂 Project Structure

Headline_Generator/
│
├── app.py                   # Main Flask application & Inference Logic
├── requirements.txt         # Dependencies
├── README.md                # Project Documentation
│
├── models/                  # Trained AI artifacts
│   ├── encoder_model.keras  # The Reader (Bidirectional GRU)
│   ├── decoder_model.keras  # The Writer (LSTM + Attention)
│   └── tokenizer.pkl        # Word-to-Integer Dictionary
│
├── static/
│   └── style.css            # Styling for the web interface
│
└── templates/
    └── index.html           # Frontend HTML structure

🧠 Model Architecture

The core of this project is a Seq2Seq model designed to map long sequences (articles) to short sequences (headlines).

1. Encoder: A Bidirectional GRU layer that processes the input text and creates a "Context Vector."
2. Attention Layer: A custom-built BahdanauAttention layer that allows the Decoder to "look back" at specific words in the article at every step of generation.
3. Decoder: An LSTM layer that predicts the headline one word at a time, conditioned on the previous word and the Attention context.

💻 Installation & Setup

1. Clone the Repository

git clone https://github.com/Ritee17/Headline_Generator
cd Headline_Generator

2. Set Up Virtual Environment

# Windows
python -m venv venv
venv\Scripts\activate

# Mac/Linux
python3 -m venv venv
source venv/bin/activate

3. Install Dependencies

pip install flask tensorflow pandas numpy

4. Run the Application

python app.py

You will see a message: Running on http://127.0.0.1:5000. Open this link in your browser.

🚧 Engineering Challenges Solved

• OOM (Out of Memory) Errors: The initial model with a 30,000-word vocabulary exceeded the 4GB VRAM limit of the RTX 3050. This was solved by implementing Mixed Precision (FP16) training and optimizing batch sizes.
• Inference Loop: Unlike training (which uses Teacher Forcing), real-world prediction requires the model to feed its own output back as input. I engineered a custom decode_sequence function to handle this recursive state management.
• Custom Layer Serialization: Flask cannot natively load custom Keras layers. I implemented a wrapper to inject the BahdanauAttention class definition during the model loading process.

👤 Author

Ritee

• Role: AI Engineer & Full-Stack Developer
• Focus: Deep Learning, NLP, System Architecture