LLM Wrapper

A comprehensive Python wrapper for Azure OpenAI with built-in PostgreSQL integration and usage tracking. Provides detailed analytics for LLM usage with support for both text and JSON response formats.

Features

• 🚀 Easy Integration: Simple API for interacting with Azure OpenAI services
• 📊 Usage Tracking: Comprehensive logging and analytics for all LLM requests
• 💾 PostgreSQL Integration: Built-in PostgreSQL database support with automatic table creation
• ⚡ High Performance: Optimized for concurrent requests and high throughput
• 🔒 Secure: Built-in security features and API key management
• 📈 Analytics: Detailed usage statistics and reporting
• 🎯 Response Types: Support for both text and JSON response formats
• 🐳 Production Ready: Robust error handling and logging

Installation

Basic Installation

pip install llm_wrapper_biz

Quick Start

Basic Usage

from llm_wrapper_biz import LLMWrapper

# Initialize the wrapper (database connection is automatic)
wrapper = LLMWrapper(
    service_url="https://your-azure-openai-instance.openai.azure.com",
    api_key="your-azure-openai-api-key",
    deployment_name="your-deployment-name",
    api_version="2023-05-15",
    default_model='gpt-4'
)

# Send a text request
response = wrapper.send_request(
    input_text="What are the benefits of renewable energy?",
    customer_id=1,
    organization_id=1,
    response_type="text",  # "text" or "json"
    temperature=0.7,
    max_tokens=2000
)

print(f"Response: {response['processed_output']}")
print(f"Tokens used: {response['total_tokens']}")
print(f"Response type: {response['response_type']}")

# Send a JSON request
json_response = wrapper.send_request(
    input_text="Create a JSON object with information about Python programming including name, creator, and year_created.",
    customer_id=1,
    organization_id=1,
    response_type="json"
)

print(f"JSON Response: {json_response['processed_output']}")
print(f"Creator: {json_response['processed_output'].get('creator', 'N/A')}")

# Get usage statistics
stats = wrapper.get_usage_stats()
print(f"Total requests: {stats['total_requests']}")
print(f"Total tokens: {stats['total_tokens']}")

# Clean up
wrapper.close()

Simplified Usage

For easier integration, use the simplified methods:

# Get just the processed output (text)
text_result = wrapper.send_request(
    input_text="Explain quantum computing",
    customer_id=1,
    organization_id=1,
    response_type="text"
)
print(text_result)

# Get just the processed output (JSON)
json_result = wrapper.send_request(
    input_text="Create JSON with weather data for London",
    customer_id=1,
    organization_id=1,
    response_type="json"
)

Response Types

Text Response (Default)

response = wrapper.send_request(
    input_text="Explain artificial intelligence",
    customer_id=1,
    organization_id=1,
    response_type="text"
)

# Response structure:
{
    "output_text": "raw response from API",
    "processed_output": "same as output_text for text responses",
    "response_type": "text",
    "input_tokens": 10,
    "output_tokens": 150,
    "total_tokens": 160,
    "response_time_ms": 1200,
    "model": "gpt-4",
    "full_response": {...}
}

JSON Response

response = wrapper.send_request(
    input_text="Create a JSON object with user information including name, age, and skills array",
    customer_id=1,
    organization_id=1,
    response_type="json"
)

# Response structure:
{
    "output_text": '{"name": "John", "age": 30, "skills": ["Python", "AI"]}',
    "processed_output": {"name": "John", "age": 30, "skills": ["Python", "AI"]},
    "response_type": "json",
    "input_tokens": 15,
    "output_tokens": 25,
    "total_tokens": 40,
    "response_time_ms": 1500,
    "model": "gpt-4",
    "full_response": {...}
}

Database Schema

The wrapper automatically creates the following PostgreSQL table:

CREATE TABLE token_usage_log (
    id SERIAL PRIMARY KEY,
    customer_id INTEGER NOT NULL,
    organization_id INTEGER NOT NULL,
    model_name VARCHAR(255) NOT NULL,
    request_params JSON,
    response_params JSON,
    input_tokens INTEGER NOT NULL,
    output_tokens INTEGER NOT NULL,
    total_tokens INTEGER NOT NULL,
    request_timestamp TIMESTAMP DEFAULT NOW(),
    response_time_ms INTEGER NOT NULL,
    status VARCHAR(50) DEFAULT 'success'
);

Usage Analytics

# Get overall statistics
stats = wrapper.get_usage_stats()

# Get customer-specific statistics
customer_stats = wrapper.get_usage_stats(customer_id=1)

# Get organization-specific statistics
org_stats = wrapper.get_usage_stats(organization_id=1)

# Get statistics for a specific time period
period_stats = wrapper.get_usage_stats(
    start_date="2024-01-01T00:00:00",
    end_date="2024-01-31T23:59:59"
)

# Example stats output:
{
    "total_requests": 150,
    "total_tokens": 45000,
    "models": [
        {
            "model_name": "gpt-4",
            "requests": 100,
            "input_tokens": 15000,
            "output_tokens": 20000,
            "total_tokens": 35000,
            "avg_response_time_ms": 1200
        },
        {
            "model_name": "gpt-3.5-turbo",
            "requests": 50,
            "input_tokens": 5000,
            "output_tokens": 5000,
            "total_tokens": 10000,
            "avg_response_time_ms": 800
        }
    ]
}

Configuration Options

Parameter	Type	Default	Description
service_url	str	Required	Azure OpenAI service endpoint URL
api_key	str	Required	Azure OpenAI API key
deployment_name	str	Required	Azure OpenAI deployment name
api_version	str	Required	Azure OpenAI API version
default_model	str	'gpt-4'	Default model identifier
timeout	int	30	Request timeout in seconds

API Reference

Core Methods

send_request(input_text, customer_id, organization_id, response_type="text", **kwargs)

Send a request to the Azure OpenAI service.

Parameters:

• input_text (str): The prompt text
• customer_id (int): Customer identifier
• organization_id (int): Organization identifier
• response_type (str): Response format - "text" or "json"
• model (str, optional): Model to use for this request
• temperature (float, optional): Sampling temperature (0.0-1.0)
• max_tokens (int, optional): Maximum tokens in response

Returns:

• dict: Response containing output text, processed output, token counts, and metadata

send_request_simple(input_text, customer_id, organization_id, response_type="text", **kwargs)

Simplified method that returns only the processed output.

Parameters:

• Same as send_request()

Returns:

• str (for text) or dict (for JSON): Direct processed output

get_usage_stats(**filters)

Get usage statistics with optional filtering.

Parameters:

• customer_id (int, optional): Filter by customer
• organization_id (int, optional): Filter by organization
• start_date (str, optional): Start date in ISO format
• end_date (str, optional): End date in ISO format

Returns:

• dict: Usage statistics including request counts, token usage, and performance metrics

close()

Close database connections and clean up resources.

Requirements

• Python 3.8+

License

This project is licensed under the MIT License.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Acknowledgments

• Thanks to all contributors who have helped shape this project
• Built with love for the AI/ML community