Enterprise Agentic Data Analyst

An asynchronous multi-agent platform for data analysis, cleaning, mutation, visualization, and modeling. Powered by LangGraph, Django Ninja, and Chainlit.

Architecture

The platform uses an orchestrator-worker topology to safely execute agent-generated code inside isolated sandboxes:

• Orchestrator Node: Uses Gemini 2.5 Flash to evaluate the user query against the dataset schema and generate a structured execution plan.
• Worker Node: Generates Python code via a local Qwen 2.5 Coder model.
• Docker Sandbox: Executes generated code in an isolated Docker container, returning console logs, exit codes, and output artifacts.
• Data Validation Loop: Evaluates data quality metrics post-execution (e.g., ensuring zero null values remain after imputation). If validation fails, it triggers self-healing.
• Session State Persistence: Stores conversation logs and completed tasks inside dataset-specific folders, allowing multi-turn operations.
• Resilient Model Strategy: Automatically rotates multiple Gemini API keys. If all keys hit rate limits or fail, it falls back to a local Ollama instance running Qwen 2.5 Coder.

Running with Docker Compose

Running the entire stack with Docker Compose is the recommended way to set up the system.

Configuration

Create a .env file in the root directory:

GEMINI_API_KEY=your_key_1,your_key_2
SECRET_KEY=django-insecure-local-mvp-key-change-this-in-production
DEBUG=True

Launching the Stack

1. Start the containers:

   docker-compose up --build

2. Pull the required model on the Ollama container:

   docker exec -it da-agent-ollama-1 ollama pull qwen2.5-coder:1.5b

3. Access the Chainlit user interface at http://localhost:8001. The Django backend is running at http://localhost:8000.

Local Development Setup

Prerequisites

• Python 3.12+
• Docker (with the daemon running)
• Ollama running Qwen 2.5 Coder 1.5B (qwen2.5-coder:1.5b via ollama pull qwen2.5-coder:1.5b)

Installation

1. Initialize the virtual environment:

   python -m venv .venv
   source .venv/bin/activate
   pip install -r req.txt

2. Run database migrations:

   python manage.py migrate

Running the Application

1. Start the Django backend server:

   python manage.py runserver 8000

2. Start the Chainlit frontend chat server:

   chainlit run frontend/app.py --port 8001

3. Open http://localhost:8001 in your browser.