AI Poetry Generator

A machine learning project that generates creative poetry using natural language processing techniques.

Overview

Developed an AI-powered poetry generator that creates original poems based on various styles and themes. The system uses advanced NLP techniques and machine learning models to understand poetic patterns and generate coherent, creative verses.

Key Features

• Multiple Poetry Styles: Generate poems in various styles (haiku, sonnet, free verse, etc.)
• Theme-Based Generation: Create poems based on specified themes or emotions
• Style Transfer: Apply the style of famous poets to new content
• Interactive Interface: Simple Python CLI for easy interaction
• Training Pipeline: Custom training system for fine-tuning on poetry datasets

Tech Stack

• Language: Python 3.9+
• ML Framework: TensorFlow/PyTorch
• NLP: spaCy, NLTK
• Models: Transformer-based language models
• Data: Curated poetry dataset (public domain works)

Technical Highlights

• Implemented sequence-to-sequence model for text generation
• Developed custom tokenization for poetic meter and rhyme scheme
• Created evaluation metrics for poetic quality (rhythm, rhyme, coherence)
• Built data preprocessing pipeline for poetry corpus
• Integrated attention mechanisms for better context understanding

Results

• Generated hundreds of unique poems across multiple styles
• Achieved coherent rhyme schemes and meter patterns
• Demonstrated understanding of NLP and generative AI
• Created extensible architecture for future enhancements

Challenges & Solutions

Challenge: Maintaining poetic structure while ensuring semantic coherence
Solution: Implemented constrained generation with rhyme and meter scoring

Challenge: Limited training data for specific poetry styles
Solution: Used transfer learning from pre-trained language models with fine-tuning

Challenge: Evaluating quality of generated poetry
Solution: Developed multi-metric evaluation system combining automated metrics with human feedback