Architecture

Documentation technique de l'architecture de Duckling.

Vue d'ensemble

Duckling is a full-stack web application with a clear separation between frontend et backend:

graph LR
    A[Navigateur] --> B[Frontend React]
    B --> C[Backend Flask]
    C --> D[Moteur Docling]
    D --> E[(Stockage)]

    style A fill:#3b82f6,color:#fff
    style B fill:#1e3a5f,color:#fff
    style C fill:#14b8a6,color:#fff
    style D fill:#8b5cf6,color:#fff
    style E fill:#f59e0b,color:#fff

Sections

Décisions de conception clés

Séparation des préoccupations

• Frontend: React avec TypeScript pour la sécurité des types et une UI moderne
• Backend: Flask pour la simplicité et l'accès à l'écosystème Python
• Moteur: Docling pour la conversion de documents (bibliothèque IBM)

Async Traitement

Document conversion is hetled asynchronously:

1. Client uploads file
2. Server returns job ID immediately
3. Client polls for status
4. Server processes in background thread
5. Results available when complete

Job Queue

A thread-based job queue prevents memory exhaustion:

• Maximum 2 concurrent conversions
• Jobs queued when capacity reached
• Automatic cleanup of completed jobs

Paramètres Persistence

Paramètres are stored per-user session et applied per-conversion:

• Global defaults in config.py
• User settings stored in database (per session ID)
• Per-request overrides via API

Paramètres are isolated per user session, ensuring multi-user deployments don't interfere with each other's preferences.

Technology Stack

Frontend

Backend