Architecture

DashAI is a modular, extensible platform for machine learning workflows. It provides a web-based interface for training models, exploring datasets, explaining predictions, and more. This document describes how DashAI works internally.

High-Level Overview

DashAI follows a client-server architecture with three main runtime processes:

1. FastAPI backend — serves the REST API on port 8000. In production it also serves the compiled React SPA at /app/.
2. Huey consumer — a background worker that processes long-running jobs (training, exploration, prediction, etc.).
3. React frontend — a single-page application that communicates with the backend through the REST API. In development it runs separately on port 3000 (yarn start); in production it is compiled and served by FastAPI.

The entry point is DashAI/__main__.py, which uses Typer as CLI. On startup it:

1. Resolves the local data path (defaults to ~/.DashAI).
2. Starts the Huey consumer. In development (normal Python install) this is an external subprocess; in bundled mode (PyInstaller) it runs as an in-process thread.
3. Starts the FastAPI server via Uvicorn.
4. Optionally opens a browser or a PyWebView window.

Dependency injection is handled by Kink. The DI container (back/container.py) wires together the database engine, session factory, component registry, and job queue so that API endpoints can receive them automatically.

Key Patterns

• Component-based extensibility — all ML functionality (models, tasks, metrics, explorers, etc.) is encapsulated in components that share a common registration and configuration mechanism.
• Schema-driven configuration — components declare Pydantic schemas that are converted to JSON Schema for dynamic frontend form generation and backend validation.
• Asynchronous job processing — long-running operations are offloaded to a Huey background worker, with status tracking via signals and database updates.
• Clean separation of concerns — the API layer handles HTTP, the component registry handles discovery, the job queue handles execution, and the database handles persistence.

Further Reading

Topic	Page
REST API structure, router map, dependency injection	API
Component types, registry, and configurable objects	Components
SQLite schema, ORM tables, and data storage	Database
Huey job queue and job types	Job System
Notebook sessions, explorers, and converters	Notebook
End-to-end training and exploration walkthroughs	Workflow Examples
Column semantic types and inference	Semantic Types