Exploring Prompt Patterns in AI-Assisted Code Generation

This project analyzes different prompt patterns in AI-assisted code generation to optimize developer-AI interactions. By evaluating 20,000+ AI-generated code responses, we identify the most effective prompting strategies that yield high-quality outputs with minimal interactions.

Tech Stack

• Python (Data processing, analysis)
• Pandas, NumPy (Data handling)
• TextBlob (Sentiment analysis)
• Matplotlib, Seaborn (Data visualization)

Key Features

• Assessed 7 Developer GPT Prompt Patterns to enhance AI-generated code quality.
• Automated classification of 20,000 AI-generated code responses using keyword algorithms.
  {Citation: S. Dicuffa. Ssw625. https://github.com/sophiadicuffa/SSW625, 2024.GitHub Repository.}
• Computed effectiveness scores based on response length, token ratio, and sentiment analysis.
• Identified the most efficient prompt structures that improve AI interaction and reduce iterative corrections.

Results

• "Context and Instruction" & "Recipe" patterns were found to be the most effective.
• Structured prompts reduced iterative interactions by 50%, improving developer productivity.
• The study provides insights into optimal prompt engineering strategies for AI-assisted coding.