🕊️ Roman Urdu Poetry Generator

AI-Powered Poetic Expression in Roman Urdu

🌸 “Where code meets emotion, and verses are born from data.” 🌸

---

---

🌟 Overview

Roman Urdu Poetry Generator is an AI-based text generator that creates poetic verses in Roman Urdu, inspired by the works of poets like Ahmad Faraz and Mir Taqi Mir.

Built using LSTM (Long Short-Term Memory) networks in PyTorch, the model learns from a curated dataset of Urdu poetry transliterated into Roman script. It then composes new, coherent poetic lines — often surprisingly soulful. 💫

You can interact with it through a Gradio Web App, adjusting:

• 🎨 Creativity (temperature)
• ✍️ Poem length (number of words)
• 📏 Structure (words per line)

---

🧠 How It Works

1. The model reads your starting phrase (e.g., “aañkh se”).
2. It predicts the next most likely word, one token at a time.
3. Words are grouped into lines, forming rhythmic verses.
4. You control how “creative” or “logical” the poem sounds with the temperature slider.

🪄 A higher temperature = more creativity & randomness. A lower one = more structure & logic.

---

⚙️ Tech Stack

Component	Technology
🧩 Model Type	LSTM (Long Short-Term Memory Network)
🧠 Framework	PyTorch
💬 Interface	Gradio
📁 Dataset	Roman Urdu Poetry.csv
🧾 Language	Python 3.9+

---

🧩 File Structure

📂 Roman-Urdu-Poetry-Generator
│
├── 📄 app.py                 # Main app (model + Gradio UI)
├── 📄 poetry_model.pth       # Trained LSTM model checkpoint
├── 📄 Roman-Urdu-Poetry.csv  # Training dataset (poems in Roman Urdu)
└── 📄 README.md              # You're reading this :)

---

🚀 Getting Started

🔧 1. Clone the Repository

git clone https://github.com/your-username/Roman-Urdu-Poetry-Generator.git
cd Roman-Urdu-Poetry-Generator

📦 2. Install Dependencies

pip install torch gradio numpy

💡 3. Run the App

python app.py

This will automatically launch the Gradio interface in your browser. 🌐

---

💻 Gradio Interface Preview

🎤 Roman Urdu Poetry Generator

Generate poetic verses in Roman Urdu using an AI trained on classic poetry.

Input	Control
🪶 Starting Word	“aañkh”, “dil”, “ishq”, etc.
🔢 Number of Words	Choose total poem length
🔥 Creativity (Temperature)	Adjust imagination level
📏 Words per Line	Control verse structure

---

🪶 Example Output

Input:

Starting Word: aañkh Number of Words: 25 Temperature: 1.0 Words per Line: 4

Generated Poetry:

aañkh se duur na ho dil se utar jaayega  
waqt ka kya hai guzarta hai guzar jaayega  
itna maanus na ho khalvat-e-gham se apni  
tu kabhi khud ko bhi dekhega to dar jaayega

💫 Sometimes, even AI has feelings.

---

🧩 Model Details

The model uses a multi-layer LSTM network to learn sequential word patterns from poetry.

Model Configuration:

• Embedding Dimension: 128
• Hidden Dimension: 256
• LSTM Layers: 2
• Dropout: 0.2
• Optimized for Roman Urdu poetic syntax

---

📚 Dataset

The dataset Roman-Urdu-Poetry.csv contains handpicked poems from various Urdu poets transliterated into Roman Urdu. Each record includes:

• Poet’s Name
• Original Roman Urdu Poem Text

🧾 Example entry:

Poet: ahmad-faraz  
Poetry: "aañkh se duur na ho dil se utar jaayega..."

---

🎨 Future Enhancements

✅ Add multilingual support (Urdu script + English translations) ✅ Train on larger poetry datasets (Mir, Faiz, Iqbal) ✅ Implement rhyme & rhythm detection ✅ Add voice output using TTS (Text-to-Speech)

---

🤝 Contributing

Contributions are always welcome! 💬 If you’d like to enhance the model, improve training data, or design a better interface, feel free to:

fork -> code -> commit -> pull request

---

🪙 License

This project is licensed under the MIT License — free to use, modify, and share.

See the LICENSE file for details.

---

✨ “Poetry is not written by code — it’s awakened by it.” ✨

Made with ❤️ using Python, PyTorch & Gradio

---