📊 Data Analyst Agent ⚡️

A small Chainlit-based data analyst chat agent that answers questions about a synthetic business dataset, runs pandas / matplotlib code when needed, and returns outputs (stream) and inline PNG plots.

OpenAI Code Interpreter

• Released in March 2023 (alpha for ChatGPT Plus).
• Integrated with Assistant API.
• Assistant API will be deprecated in 2026 ⚠️; use Responses API instead.
• Responses API creates an isolated container to execute Python code, with outputs such as images stored within the container.
• Open-source alternatives (e.g., Open Interpreter) are not actively developed; using Responses API is recommended. Integration with Chainlit has issues:
  1. Open Interpreter (v0.4.3) sometimes fails to install with Chainlit (v2.7.2) due to dependency collisions.
  2. Code Interpreter API (v0.1.20) has issues with Azure OpenAI.

Synthetic data description

• synthetic_business_data.csv

  • Synthetic, tabular business / transaction data for demo and analysis.

• synthetic_metadata.json

  • JSON object describing columns and their human-readable descriptions.

Usage

1. Create a .env with required Azure variables:
   • AZURE_OPENAI_ENDPOINT
   • AZURE_OPENAI_DEPLOYMENT_NAME
   • AZURE_OPENAI_API_KEY
   • AZURE_OPENAI_API_VERSION
2. Install dependencies:
   • poetry install
3. Run the app:
   • python app.py
4. In the chat, ask questions about the dataset. The assistant will run code for calculations/plots and return results.

Notes

• Tool execution writes a Python script, runs it, and returns stdout/stderr plus any generated PNGs inline.
• File uploads are disabled in the chat flow.
• app.legacy.py is present in the repo but is considered legacy / not used by the code.
• If you plan to deploy or extend for Azure, follow Azure best practices, secure .env secrets 🔐, and review deployment guidance.