Neural Network Poetry Generator

Current implementation: frontend demonstration that selects from curated poem templates and calculates simple lexical statistics. It does not currently call or train a transformer model; model-driven generation remains part of the roadmap.

This project explores the intersection of artificial intelligence and creative writing by training neural networks on classical poetry collections to generate new, original verse. The interface allows users to experiment with different prompts, styles, and parameters to understand how AI interprets and recreates poetic forms.

{% MediaEmbed platform=“youtube” url=“https://www.youtube.com/watch?v=dQw4w9WgXcQ” title=“Demo: AI Poetry Generation in Action” description=“Watch how the neural network creates poetry in real-time, responding to user prompts and style preferences.” width=560 height=315 autoplay=false controls=true /%}

Technical Approach

The system uses a fine-tuned GPT model trained on a curated dataset of poetry from the 18th and 19th centuries. The training process focused on:

• Meter and rhythm patterns
• Rhyme scheme recognition
• Thematic coherence
• Historical language patterns

Interactive Features

• Theme Selection: Choose from predefined themes or input custom prompts
• Style Controls: Adjust formality, complexity, and historical period
• Real-time Generation: Watch the AI compose verse line by line
• Analysis Mode: View the model’s confidence and decision-making process

Educational Impact

This tool has been used in digital humanities courses to help students understand both the possibilities and limitations of AI in creative domains. It raises important questions about authorship, creativity, and the role of technology in artistic expression.