fAIr Documentation

Appearance

Project Architecture ​

This page provides a high-level overview of the fAIr platform’s architecture, including its main components, data flow, and technology stack.

Overview ​

fAIr is a modular, containerized platform designed for AI-assisted mapping. It consists of several services that work together to process imagery, train AI models, and assist users in mapping tasks.

Main Components ​

  • Frontend
    Built with React and TypeScript, the frontend provides an intuitive web interface for users to manage datasets, train models, and perform mapping tasks.

  • Backend API
    Powered by Django (Python), the backend exposes REST APIs for user management, dataset handling, model training, and prediction services.

  • AI/ML Workers
    Use TensorFlow and PyTorch for training and running AI models. Workers are managed via Celery for distributed task processing.

  • Database
    PostgreSQL with PostGIS extension stores user data, project metadata, and geospatial information.

  • Redis
    Used as a message broker for Celery task queues.

  • Flower
    Provides a web UI for monitoring Celery tasks and workers.

  • Developer Documentation
    Served via VitePress. Run npm run docs:dev and visit the port shown in your terminal.

Data Flow ​

  1. User uploads or selects imagery (e.g., from OpenAerialMap).
  2. Frontend sends requests to the backend API to create datasets and manage projects.
  3. Backend schedules AI tasks (training, prediction) via Celery.
  4. AI/ML workers process imagery and labels, train models, and generate predictions.
  5. Results are stored in the database and made available to the frontend for review and editing.
  6. Users review, edit, and validate predictions before exporting or uploading to OSM.

Technology Stack ​

  • Frontend: React, TypeScript, Vite
  • Backend: Django, Django REST Framework
  • AI/ML: TensorFlow, PyTorch
  • Task Queue: Celery, Redis
  • Database: PostgreSQL, PostGIS
  • Containerization: Docker, Docker Compose
  • Documentation: MkDocs Material

Deployment ​

All services are containerized and orchestrated using Docker Compose.
See Docker Installation (Linux/Mac) or Docker Installation (Windows) for setup instructions.

Extending the Platform ​

  • Add new AI models: Integrate custom models by extending the AI worker codebase.
  • API integrations: Use the REST API to connect with external tools or automate workflows.
  • Frontend plugins: Enhance the user interface with new features or visualization tools.

Diagram ​

A system architecture diagram can be added here for visual reference (e.g., using diagrams.net or Mermaid).

For more details, see the contributing guide or open an issue on GitHub.