core4ai 1.1.6

Contextual Optimization and Refinement Engine for AI

Core4AI: Contextual Optimization and Refinement Engine for AI

Core4AI is an intelligent system that transforms basic user queries into optimized prompts for AI systems using MLflow Prompt Registry. It dynamically matches user requests to the most appropriate prompt template and applies it with extracted parameters.

Architecture

Core4AI's architecture is designed for seamless integration with MLflow while providing flexibility in AI provider selection:

This integration allows Core4AI to leverage MLflow's tracking capabilities for prompt versioning while providing a unified interface to multiple AI providers.

✨ Key Features

• 📚 Centralized Prompt Management: Store, version, and track prompts in MLflow
• 🧠 Intelligent Prompt Matching: Automatically match user queries to optimal templates
• 🔄 Dynamic Parameter Extraction: Identify and extract parameters from natural language
• 🔍 Content Type Detection: Recognize the type of content being requested
• 🛠️ Multiple AI Providers: Seamless integration with OpenAI and Ollama
• 📊 Detailed Response Tracing: Track prompt optimization and transformation stages
• 📝 Version Control: Track prompt history with production and archive aliases
• 🧩 Extensible Framework: Add new prompt types without code changes

🚀 Installation

Basic Installation

# Install from PyPI
pip install core4ai

Development Installation

# Clone the repository
git clone https://github.com/iRahulPandey/core4ai.git
cd core4ai

# Install in development mode
pip install -e ".[dev]"

⚙️ Initial Configuration

Core4AI can be configured either through the CLI setup wizard or programmatically via the Python API.

CLI Setup (Recommended for First-Time Setup)

# Run the setup wizard
core4ai setup

The wizard will guide you through:

1. MLflow Configuration: Enter the URI of your MLflow server (default: http://localhost:8080)
2. Existing Prompts Import: Import existing prompts from MLflow if needed
3. AI Provider Selection: Choose between OpenAI or Ollama and configure the selected provider

Python API Setup

from core4ai import Core4AI

# Create a Core4AI instance
ai = Core4AI()

# Set MLflow URI
ai.set_mlflow_uri("http://localhost:8080")

# Configure OpenAI
api_key = os.environ.get("OPENAI_API_KEY")  # Get from environment
ai.configure_openai(api_key=api_key, model="gpt-3.5-turbo")

# Or configure Ollama
# ai.configure_ollama(uri="http://localhost:11434", model="llama3.2:latest")

# Save the configuration
ai.save_config()

📝 Prompt Management

Core4AI uses a powerful prompt management system that allows you to create, register, and use prompt templates in various formats.

Prompt Template Format

Core4AI uses markdown files to define prompt templates. Each template should follow this structure:

# Prompt Name: example_prompt

## Description
A brief description of what this prompt does.

## Tags
type: example
task: writing
purpose: demonstration

## Template
Write a {{ style }} response about {{ topic }} that includes:
- Important point 1
- Important point 2
- Important point 3

Please ensure the tone is {{ tone }} and suitable for {{ audience }}.

Key Guidelines

1. Prompt Name is required and must:

   • Be the first line of the file
   • Use the format # Prompt Name: name_prompt
   • End with _prompt suffix
   • Use underscores instead of spaces (e.g., cover_letter_prompt)

2. Template Section must:

   • Use double braces for variables: {{ variable_name }}
   • Have at least one variable
   • Provide clear instructions

3. Tags Section is recommended and should include:

   • type: The prompt category (e.g., essay, email, code)
   • task: The purpose (e.g., writing, analysis, instruction)
   • Additional metadata as needed

Creating Prompt Templates

CLI Approach

# Create a new prompt template in the current directory
core4ai register --create email

# Create a prompt template in a specific directory
core4ai register --create blog --dir ./my_prompts

Python API Approach

from core4ai import Core4AI

ai = Core4AI()

# Create a new prompt template in the current directory
result = ai.create_prompt_template("email")

# Create a prompt template in a specific directory
result = ai.create_prompt_template("blog", output_dir="./my_prompts")

This will:

1. Create a template file with the proper structure
2. Open it in your default editor for customization
3. Offer to register it immediately after editing

Registering Prompts

CLI Approach

# Register a single prompt directly
core4ai register --name "email_prompt" "Write a {{ formality }} email..."

# Register from a markdown file
core4ai register --markdown ./my_prompts/email_prompt.md

# Register all prompts from a directory
core4ai register --dir ./my_prompts

# Register built-in sample prompts
core4ai register --samples

# Register only prompts that don't exist yet
core4ai register --dir ./my_prompts --only-new

Python API Approach

from core4ai import Core4AI

ai = Core4AI()

# Register a single prompt directly
ai.register_prompt(
    name="email_prompt",
    template="Write a {{ formality }} email...",
    tags={"type": "email", "task": "writing"}
)

# Register from a markdown file
ai.register_from_markdown("./my_prompts/email_prompt.md")

# Register built-in sample prompts
ai.register_samples()

# Register from a JSON file
ai.register_from_file("./my_prompts.json")

Managing Prompt Types

Core4AI automatically tracks prompt types based on the prompt names:

CLI Approach

# List all registered prompt types
core4ai list-types

Python API Approach

from core4ai import Core4AI

ai = Core4AI()

# List all registered prompt types
prompt_types = ai.list_prompt_types()
print(prompt_types)

# Add a new prompt type
ai.add_prompt_type("custom_type")

The type is extracted from the prompt name:

• For email_prompt → type is email
• For cover_letter_prompt → type is cover_letter

Listing Available Prompts

CLI Approach

# List all prompts
core4ai list

# Show detailed information
core4ai list --details

# Get details for a specific prompt
core4ai list --name email_prompt@production

Python API Approach

from core4ai import Core4AI

ai = Core4AI()

# List all prompts
prompts = ai.list_prompts()
print(prompts)

# Get the configuration
config = ai.get_current_config()
print(config)

🛠️ Using Core4AI

Basic Chat Interactions

CLI Approach

# Simple query - Core4AI will match to the best prompt template
core4ai chat "Write about the future of AI"

# Get a simple response without enhancement details
core4ai chat --simple "Write an essay about climate change"

# See verbose output with prompt enhancement details
core4ai chat --verbose "Write an email to my boss about a vacation request"

Python API Approach

from core4ai import Core4AI

ai = Core4AI()

# Simple query
response = ai.chat("Write about the future of AI")
print(response["response"])

# With verbose output
response = ai.chat("Write an email to my boss about a vacation request", verbose=True)
print(response["prompt_match"])  # Show matched prompt details
print(response["enhanced_query"])  # Show enhanced query
print(response["response"])  # Show final response

Sample Prompts

Core4AI comes with several pre-registered prompt templates:

CLI Approach

# Register sample prompts
core4ai register --samples

Python API Approach

from core4ai import Core4AI

ai = Core4AI()

# Register sample prompts
result = ai.register_samples()
print(f"Registered {result.get('registered', 0)} prompts")

This will register the following prompt types:

Prompt Type	Description	Sample Variables
essay_prompt	Academic writing	topic
email_prompt	Email composition	formality, recipient_type, topic, tone
technical_prompt	Technical explanations	topic, audience
creative_prompt	Creative writing	genre, topic
code_prompt	Code generation	language, task, requirements
cover_letter_prompt	Cover letter writing	position, company, experience_years
qa_prompt	Question answering	topic, tone, formality
tutorial_prompt	Step-by-step guides	level, task, tool_or_method
marketing_prompt	Marketing content	content_format, product_or_service, target_audience
report_prompt	Report generation	length, report_type, topic
social_media_prompt	Social media posts	number, platform, topic
data_analysis_prompt	Data analysis reports	data_type, subject, data
comparison_prompt	Compare items or concepts	item1, item2
product_description_prompt	Product descriptions	length, product_name, product_category
summary_prompt	Content summarization	length, content_type, content
test_prompt	Test examples	formality

Each prompt is designed for specific use cases and includes variables that can be automatically extracted from user queries. You can view the details of any prompt with:

CLI Approach

# View details of a specific prompt
core4ai list --name essay_prompt@production --details

Python API Approach

from core4ai.prompt_manager.registry import get_prompt_details

# Get details of a specific prompt
details = get_prompt_details("essay_prompt@production")
print(details)

🔄 Provider Configuration

OpenAI

CLI Approach

# Set environment variable (recommended)
export OPENAI_API_KEY="your-api-key-here"

# Or configure during setup
core4ai setup
# Choose OpenAI and follow the prompts

Python API Approach

from core4ai import Core4AI
import os

ai = Core4AI()

# Using environment variable (recommended)
ai.configure_openai(model="gpt-3.5-turbo")

# Or with explicit API key (less secure)
api_key = "your-api-key-here"
ai.configure_openai(api_key=api_key, model="gpt-4")

# Save configuration (API key won't be saved to disk)
ai.save_config()

Available models include:

• gpt-3.5-turbo (default)
• gpt-4
• gpt-4-turbo
• gpt-4o

Ollama

CLI Approach

# Install and start Ollama
ollama serve

# Configure Core4AI
core4ai setup
# Choose Ollama and follow the prompts

Python API Approach

from core4ai import Core4AI

ai = Core4AI()

# Configure to use Ollama
ai.configure_ollama(uri="http://localhost:11434", model="llama2")

# Save configuration
ai.save_config()

📋 Command Reference

Command	Description	CLI Example	Python API Example
Setup	Configure Core4AI	core4ai setup	ai = Core4AI() ai.set_mlflow_uri("http://localhost:8080") ai.configure_openai()
Register	Register prompts	core4ai register --samples	ai.register_samples()
List	List available prompts	core4ai list --details	ai.list_prompts()
List Types	List prompt types	core4ai list-types	ai.list_prompt_types()
Chat	Chat with enhanced prompts	core4ai chat "Write about AI"	response = ai.chat("Write about AI") print(response["response"])
Version	Show version info	core4ai version	from core4ai import __version__ print(__version__)

📊 How Core4AI Works

Core4AI follows this workflow to process queries:

1. Query Analysis: Analyze the user's query to determine intent
2. Prompt Matching: Match the query to the most appropriate prompt template
3. Parameter Extraction: Extract relevant parameters from the query
4. Template Application: Apply the template with extracted parameters
5. Validation: Validate the enhanced prompt for completeness and accuracy
6. Adjustment: Adjust the prompt if validation issues are found
7. AI Response: Send the optimized prompt to the AI provider

From Query to Enhanced Response

The user experience with Core4AI is straightforward yet powerful:

This workflow ensures that every user query is intelligently matched to the optimal prompt template stored in MLflow, parameters are properly extracted and applied, and the result is validated before being sent to the AI provider.

Troubleshooting Installation

NumPy Binary Incompatibility

If you encounter an error like this during installation or when running Core4AI:

ValueError: numpy.dtype size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject

Try reinstalling in the following order:

# Remove the problematic packages
pip uninstall -y numpy pandas mlflow core4ai

# Reinstall in the correct order with specific versions
pip install numpy==1.26.0
pip install pandas
pip install mlflow>=2.21.0
pip install core4ai

MLflow Server Connection Issues

If you encounter problems connecting to MLflow:

1. Make sure your MLflow server is running:

   mlflow server --host 0.0.0.0 --port 8080
   

2. Verify connection:

   curl http://localhost:8080
   

3. Configure Core4AI to use your MLflow server:

   core4ai setup
   

   Or with Python:

   from core4ai import Core4AI
   ai = Core4AI()
   ai.set_mlflow_uri("http://localhost:8080")
   ai.save_config()
   

📜 License

This project is licensed under the Apache License 2.0