Generative AI Prompt extensions

A flexible, composable library for constructing and transforming LLM prompts with a fluent interface.

Installation

pip install aipp

Development Setup

uv venv
source .venv/bin/activate
uv sync --group e
pytest -s
code .

Overview

Generative AI Prompt extensions helps you build complex prompts with:

• Chain-of-thought reasoning instructions
• Variable substitution
• Few-shot examples
• Role/persona assignment
• JSON data formatting
• Structured output requirements
• And more...

Quick Start

from aipp import PromptBuilder

# Simple example
prompt = (
    PromptBuilder()
    .configure_variables(model="GPT-4", task="summarization")
    .step_by_step(detailed=True)
    .prompt("Create a summary using {model} for {task} task.")
)

print(str(prompt))

Output:

Please provide a detailed, step-by-step explanation with thorough reasoning at each step.
Create a summary using GPT-4 for summarization task.

Features

Variable Substitution

prompt = (
    PromptBuilder()
    .configure_variables(model="GPT-4", task="summarization", length="short")
    .prompt("Create a {length} summary using {model} for {task} task.")
)

JSON Data

data = {
    "user": {"name": "Alice", "preferences": ["science", "history"]},
    "settings": {"language": "en", "format": "paragraph"}
}

prompt = (
    PromptBuilder()
    .prompt("Generate content based on user preferences")
    .with_json(data)
)

Few-Shot Examples

examples = [
    ("Summarize: The quick brown fox jumps over the lazy dog.",
     "A fox quickly jumps over a dog that is resting."),
    ("Summarize: The study showed significant results with p<0.05.",
     "The research findings were statistically significant.")
]

prompt = (
    PromptBuilder()
    .prompt("Summarize the following text")
    .with_few_shot(examples)
)

Structured Output

schema = {
    "type": "object",
    "properties": {
        "summary": {"type": "string"},
        "length": {"type": "integer"},
        "topics": {"type": "array", "items": {"type": "string"}}
    }
}

prompt = (
    PromptBuilder()
    .prompt("Analyze the following text")
    .structured_output(schema)
)

Chain of Thought

prompt = (
    PromptBuilder()
    .prompt("Solve this math problem")
    .chain_of_thought()
)

Role/Persona Assignment

prompt = (
    PromptBuilder()
    .prompt("Explain quantum computing")
    .as_role("quantum physicist", "with expertise in quantum information theory")
)

Lambda Processors

Use inline lambda functions for custom transformations:

prompt = (
    PromptBuilder()
    .prompt("Standard prompt text")
    .process(lambda content, _: f"IMPORTANT: {content}")
    .process(lambda content, _: f"{content}\n\nPlease be thorough.")
)

Combining Processors

Processors can be combined in any order:

prompt = (
    PromptBuilder()
    .prompt("Summarize the following article")
    .chain_of_thought()
    .with_few_shot(examples)
    .structured_output(schema)
    .as_role("professional writer")
    .process(lambda content, _: f"{content}\n\nLimit response to 3 paragraphs.")
)

Advanced Usage

Custom Processors

Create your own processors:

from aipp.processors.base import SimpleProcessor


def create_custom_processor(options):
    def transform(content, context=None):
        # Transform the content based on options
        return f"Custom prefix: {content}"

    return SimpleProcessor(transform, name="custom_processor")


prompt = (
    PromptBuilder()
    .prompt("Base content")
    .add_processor(create_custom_processor({"option": "value"}))
)

Processing Context

Processors can share context:

def create_metadata_processor():
    def transform(content, context=None):
        if context is not None:
            context['processed_at'] = "2023-01-01"
        return content
    
    return SimpleProcessor(transform, name="metadata")

prompt = PromptBuilder()
prompt.add_processor(create_metadata_processor())
prompt.prompt("Content")
print(prompt.metadata)  # Contains processed_at

Contributing

We welcome contributions to improve Generative AI Prompt extensions! See CONTRIBUTING.md for detailed information on:

• Architecture overview
• Core concepts
• How to extend the library
• Design patterns used
• Best practices
• Testing guidelines

Please submit issues and pull requests following our guidelines.

License

MIT