PROSE Prompt Optimizer

This project implements a local PROSE-style prompt optimizer based on the referenced paper, including:

• Flask web UI and /optimize_prompt API
• Rule-based, ML-guided, and GA-inspired candidate generation
• Evaluation metrics for semantic fidelity, clarity, readability, token efficiency, and improvement
• A synthetic 200-prompt training dataset following the paper's 8-domain setup
• A local TF-IDF + Logistic Regression predictor for domain, mode, strategy, and ambiguity

Run

python app.py

Open http://127.0.0.1:5000.

Generate Training Data

python training/generate_dataset.py

This creates data/prose_training_prompts.csv with 200 prompts across:

General, Education, Coding, Writing, Summarization, Reasoning, Technical/Data, and Business/Finance.

Train

python training/train_model.py

This creates:

• models/prose_predictor.joblib
• models/training_report.txt

The app automatically uses the trained model when models/prose_predictor.joblib exists.