Adaptive Prompt Refinement for Automated Requirements Analysis

An iterative pipeline that autonomously selects and refines the best prompts for generating structured requirements analysis documents.

🔍 Project Overview

This project explores Dynamic Adaptive Prompt Optimization (DAPO) to automatically generate, evaluate, and refine prompts for high-quality requirements analysis using Large Language Models (LLMs). The system applies various prompting techniques and leverages an LLM-as-a-Judge approach to determine the most effective prompt.

For a full research breakdown, please refer to MY-RESEARCH.md.

🚀 Key Features

• Automated Prompt Refinement: Uses an iterative pipeline to optimize prompt engineering strategies.
• LLM-as-a-Judge Evaluation: Selects the best prompt based on structured criteria.
• Comprehensive Analysis: Evaluates Zero-Shot, Few-Shot, Chain-of-Thought (CoT), and Meta-Prompting techniques.
• Structured Requirements Generation: Produces high-quality requirements documents automatically.

🔧 How It Works

The DAPO pipeline consists of four key stages:

1. Generate Prompt Variations (generator.py)
   Generates multiple prompt variations (Zero-Shot, Few-Shot, CoT, Meta-Prompting) with different temperatures and token limits.
2. Evaluate & Select Best Prompt (judge.py)
   Uses an LLM to analyze prompt outputs and select the most effective one.
3. Execute Final Best Prompt (executioner.py)
   Runs the selected "ultimate" prompt and generates the final requirements document.
4. Final Analysis (jury.py)
   Re-evaluates all outputs to ensure the best possible selection.

For a more detailed methodology, see MY-RESEARCH.md.

📥 Installation & Usage

1️⃣ Clone the Repository

Run the following commands:

• git clone https://github.com/your-username/adaptive-prompt-refinement.git
• cd adaptive-prompt-refinement

2️⃣ Install Dependencies

Run:

• pip install -r requirements.txt

3️⃣ Run the Full Pipeline

Run:

• python pipeline.py

4️⃣ View Results

Check the outputs:

• Final requirements analysis: artifacts/requirements_analysis.md
• Full experiment results: artifacts/final_analysis_report.md

📄 Research & Documentation

• MY-RESEARCH.md – Full research methodology and findings.
• Final Analysis Report – Detailed breakdown of prompt evaluations.
• Generated Requirements Analysis – Final structured requirements document.

📌 Contributors

• Evan R. Parra
• Carl Villarosa
• Academic Supervisor: Dr. Fernando Koch

📜 License

This project is licensed under the MIT License.

📌 For any inquiries, feel free to reach out or open an issue! 🚀