PromptFletcher 🚀

Deterministic Auto-Prompt Engineering for Python & NLP

Overview

PromptFletcher is a lightweight, deterministic, and dependency-minimal Python library for automatic prompt refinement using classical NLP heuristics.

Unlike LLM-based prompt optimizers, PromptFletcher:

• Does not require external APIs
• Does not use large transformer models
• Works offline
• Produces reproducible results
• Is fast enough for CI, batch jobs, and research pipelines

It is ideal for developers, researchers, and teams who want structured prompt improvement without LLM overhead.

Key Capabilities

• Heuristic-based prompt evaluation
• Iterative prompt refinement
• Context-aware relevance scoring
• Fast execution using NLTK
• Deterministic and reproducible behavior
• Minimal runtime footprint

Installation

Install from PyPI (recommended)

pip install promptfletcher

Install from GitHub (latest)

pip install git+https://github.com/Vikhram-S/PromptFletcher.git

On first use, required NLTK resources are downloaded automatically.

Quick Start

Initialize the Engine

from promptfletcher import AutoPromptEngineer

engineer = AutoPromptEngineer()

Define Context and Prompt

context = "We are optimizing prompt quality for large language models." prompt = "How improve AI responses"

Refine the Prompt

refined = engineer.refine_prompt(prompt, context) print(refined)

Example output:

How improve AI responses? Please explain in detail.

How It Works

PromptFletcher applies classical NLP heuristics to evaluate and refine prompts:

This design makes the system predictable, explainable, and easy to debug.

API Reference

AutoPromptEngineer

refine_prompt(prompt: str, context: str, iterations: int = 3) -> str
Refines a prompt iteratively using heuristic scoring.

evaluate_prompt(prompt: str, context: str) -> float
Returns a numeric quality score for a prompt.

Use Cases

• Prompt standardization in teams
• Automated prompt cleanup pipelines
• Prompt benchmarking
• Research experiments
• CI validation of prompts
• Offline prompt tuning
• Educational NLP projects

Dependencies

PromptFletcher keeps dependencies intentionally minimal:

• nltk >= 3.6
• numpy >= 1.21
• regex >= 2023.3.23

License

PromptFletcher is released under the MIT License.
You are free to use, modify, and distribute this software.

Contributing

Contributions are welcome.

Workflow:

1. Fork the repository
2. Create a branch (feature/your-feature)
3. Commit your changes
4. Push and open a Pull Request

Please ensure code is typed, documented, and non-breaking.

Support and Contact

Issues: https://github.com/Vikhram-S/PromptFletcher/issues
Email: vikhrams@saveetha.ac.in

Support the Project

If PromptFletcher helps you:

• Star the repository
• Use it in your projects
• Share it with others

Your support helps grow the project.