POP-guotai 0.3.2

Reusable, mutable, prompt functions for LLMs.

Prompt Oriented Programming (POP)

Reusable, composable prompt functions for LLMs. POP treats prompts like first-class functions: reusable, mutable, and structured for programmatic execution. It supports prompt enhancement, function/code generation, multiple LLM backends, and embeddings.

PyPI Link: https://pypi.org/project/POP-guotai/

---

Table of Contents

0. Updates

1. Features

2. Installation

3. Setup & Configuration

4. Usage

   • PromptFunction Class
   • Improving Prompts
   • Function Schema & Code Generation
   • Embeddings
   • Web Snapshot Utility

5. Example

6. Future Plans

7. Contributing

---

Updates

• 0.3.1: add image support to gemini and openai client

---

Features

• Prompt as a Function: Define reusable prompts with <<<placeholders>>> for flexible execution.

• Multi-LLM Support: Use OpenAI (default), GCP Gemini, local PyTorch, or Deepseek stubs.

• Function Schema & Code Generation:

  • Turn natural language descriptions into OpenAI function schemas.
  • Optionally generate full Python function code with docstrings and type hints.

• Prompt Improvement: Enhance base prompts using Fabric-inspired meta-prompts.

• Embeddings:

  • OpenAI and Jina API embeddings
  • Local Hugging Face model support

• Utility Functions:

  • get_text_snapshot(url) for webpage-to-text extraction with optional image captioning.

---

Installation

Install from PyPI:

pip install POP-guotai

Or from source:

git clone https://github.com/sgt1796/POP.git
cd POP
pip install -e .

---

Setup & Configuration

1. Create a .env file in your project root:

OPENAI_API_KEY=your_openai_key
GEMINI_API_KEY=your_gcp_gemini_key
JINAAI_API_KEY=your_jina_api_key

2. Dependencies are automatically handled via setup.py:

• openai, requests, python-dotenv, pydantic, transformers, numpy, backoff

---

Usage

PromptFunction Class

from POP import PromptFunction

pf = PromptFunction(
    sys_prompt="You are a helpful AI assistant.",
    prompt="Write a short poem about <<<topic>>>."
)

result = pf.execute(topic="space travel")
print(result)

---

Improving Prompts

improved_prompt = pf._improve_prompt()
print(improved_prompt)

---

Function Schema & Code Generation

# 1. Generate a JSON function schema
schema = pf.generate_schema(description="Multiply two integers and return the product.")

# 2. Generate actual Python code
code = pf.generate_code(schema)
print(code)

---

Embeddings

from POP.Embedder import Embedder

embedder = Embedder(use_api="openai")
vectors = embedder.get_embedding(["Hello world", "POP is awesome!"])
print(vectors.shape)  # (2, embedding_dim)

---

Web Snapshot Utility

from POP import get_text_snapshot

content = get_text_snapshot("https://example.com", image_caption=True)
print(content[:500])

---

Example

from POP import PromptFunction

pf = PromptFunction(
    prompt="Draw a simple ASCII art of <<<object>>>."
)

print(pf.execute(object="a cat"))
print(pf.execute(object="a rocket"))

Sample Output:

 /\_/\  
( o.o )
 > ^ <  

   /\
  /  \
 /    \
 |    |
 |    |

---

Future Plans

• Complete support for local PyTorch and Deepseek clients
• Prompt chaining and workflow composition
• Automated prompt testing framework
• Extended multimodal support (image + text prompts)

---

Contributing

1. Fork the repo
2. Create a feature branch
3. Submit a PR with clear commit messages and examples/tests

---