Undress

The Problem

Building a scalable image processing pipeline that handles AI model inference, queues jobs for async processing, and manages a credit-based billing system.

What I Built

A full-stack platform with:

• Async job queue for image processing tasks — users submit, jobs run in background, results delivered when ready
• Credit-based billing — users purchase credits, each operation deducts from balance
• Angular dashboard showing job history, credit balance, and processing status
• Admin panel for monitoring queue health and user activity

Technical Approach

• FastAPI with background task workers handling AI inference jobs
• Queue-based architecture to decouple request submission from processing — handles load spikes without dropping requests
• Angular frontend with real-time status polling for in-progress jobs
• Credit transactions with idempotency to prevent double-charging

What I’d Do Differently

Would use Redis-backed task queues (Celery or ARQ) instead of FastAPI background tasks — better reliability, retry logic, and observability for failed jobs.