Talk to Data Service

This service provides an API for querying and interacting with data using Large Language Models (LLMs) while applying differential privacy.

Getting Started

Follow these steps to set up and run the service locally.

1. Prerequisites

• Python 3.8+

2. Install Dependencies

It is recommended to use a virtual environment. Install the required dependencies using pip:

pip install -r requirements.txt

3. Environment Variables

Before starting the server, you need to configure your environment variables. Copy the .env.example file to a new file named .env:

cp .env.example .env

Set the appropriate values in your .env file. The service supports the following environment variables:

LLM Configuration:

• LLM_PROVIDER: The LLM provider to use. Supported options: openai, anthropic, azure, google.
• LLM_API_KEY: Your API key for the chosen LLM provider.
• LLM_MODEL: The specific model to use (e.g., gpt-4o, claude-3-opus-20240229, gemini-1.5-flash).
• RESPONSE_LLM_PROVIDER (Optional): Override the provider used specifically for generating the final response.
• RESPONSE_LLM_MODEL (Optional): Override the model used specifically for generating the final response.

Privacy Configuration:

• EPSILON_TOTAL: Total privacy budget available.
• EPSILON_BASE: Base privacy budget consumption.
• MIN_GROUP_SIZE: Minimum group size for data aggregation to preserve privacy.

Other:

• LOG_LEVEL (Optional): Logging level (e.g., INFO, DEBUG). Defaults to INFO.

4. Start the Server

Start the Flask server by running:

python server.py

The server will start and listen on http://0.0.0.0:8000.

API Endpoints

The service exposes the following main endpoints:

Talk to Data

• URL: /api/talk-to-data
• Method: POST
• Description: Main endpoint for asking questions about your data.
• Payload Example:

  {
      "question": "Your question here",
      "data": {},
      "ontology_url": "https://example.com/ontology",
      "sessionId": "optional-session-id",
      "privacy_mode": true
  }

Privacy Budget

• URL: /api/privacy-budget

• Method: GET

• Description: Retrieve the current remaining global privacy budget.

• URL: /api/privacy-budget/reset

• Method: POST

• Description: Reset the privacy budget (useful for testing/demo purposes).

Health Check

• URL: /api/ping
• Method: GET
• Description: Simple ping endpoint to check if the server is running.

Further Documentation

For more detailed information about the internal workings and the orchestration logic, please refer to the SERVICE.md and api.md files included in the repository.