SMIA

SMIA is a full-stack platform for managing AI projects, tracking ISO 42001 compliance, evaluating models and generating audit reports.

The repository is split into a FastAPI backend and a Vite/React front-end.

---

Repository Structure

backend/
  app/            FastAPI application
    auth.py       JWT authentication & user management
    db.py         SQLModel database setup (SQLite by default)
    main.py       API entry point, middleware and router registration
    models.py     ORM models for users, teams, projects, documents, runs, …
    routers/      Endpoints for projects, teams, documents, templates, etc.
    metrics/      Pipeline for performance, fairness and drift analysis
    tasks/        Background schedulers & cleanup jobs
    utils/        File helpers, PDF/report generation, dependencies
  config/         ISO 42001 requirements and document templates
  requirements.txt        Full API dependency list
  requirements-ml.txt     ML runtime dependencies
  Dockerfile.smia-runtime Docker image for ML tasks
  storage/        Persistent storage for uploaded data and models
  smia.db         Default SQLite database

frontend/
  src/            React + TypeScript source
    pages/        Project, team, document, dashboard pages…
    components/   Reusable UI components
    context/      Auth and team selection contexts
    api/          Centralised API client
  package.json    Web client dependencies & scripts
  vite.config.ts  Vite build configuration

README.md
.gitignore

---

Backend

The backend is a FastAPI & SQLModel application exposing a REST/JSON API.

Highlights

• Authentication: OAuth2 + JWT tokens stored in an smia_token cookie.
• Teams & Projects: CRUD operations with access control via team membership.
• ISO 42001 Checklist: Configurable controls and evidence requirements for audits.
• Document Management: Upload, version and store documents and images.
• Model Runs & Evaluations: Track training/evaluation runs, compute metrics, fairness, drift and attack robustness.
• Notifications & Dashboard: Dedicated routers for user notifications and dashboard data.
• Scheduled Tasks: Periodic cleanup of runs/logs and Docker containers.
• Report Generation: Build audit and risk-analysis reports in HTML/PDF.
• Storage Utilities: Helpers for file handling and PDF conversions.

Running the API

cd backend
python -m venv .venv
source .venv/bin/activate   # On Windows: .venv\Scripts\activate
pip install -r requirements.txt

# Optional environment variables
export DATABASE_URL="sqlite:///./smia.db"
export JWT_SECRET="change_me"   # also JWT_ALGORITHM, ACCESS_TOKEN_EXPIRE_MINUTES

uvicorn app.main:app --reload

The API listens on http://127.0.0.1:8000. Switch databases by overriding DATABASE_URL.

Optional ML Runtime

Dockerfile.smia-runtime installs heavy ML dependencies (requirements-ml.txt) in a slim image used for training/evaluation jobs.

---

Frontend

The frontend is a Vite + React + TypeScript application styled with Material UI (MUI) and Tailwind utilities.

Running the Client

cd frontend
npm install
npx vite --port 3000

The client runs on http://localhost:3000. Use the VITE_API_URL environment variable to point to a different backend URL.

Key Pages

• Login / Signup – user authentication
• Teams – browse and create teams, manage invitations
• Projects – list/create/edit AI projects with search and filtering
• Documents – upload, version and view documents & history
• Dashboard & Reports – visualise metrics, generate compliance reports
• Settings – user profile and application settings

---

Configuration

• Database: set via DATABASE_URL (defaults to SQLite file smia.db).
• JWT Security: JWT_SECRET, JWT_ALGORITHM, ACCESS_TOKEN_EXPIRE_MINUTES.
• Frontend API base: VITE_API_URL.
• Storage: uploaded data/models/logs under backend/storage/.

---

Contributing

1. Fork the repo.
2. Create a feature branch.
3. Commit and push your changes.
4. Open a pull request.