Design a Text-to-Video Generation Service
Company: OpenAI
Role: Software Engineer
Category: ML System Design
Difficulty: medium
Interview Round: Technical Screen
Design a large-scale text-to-video generation service similar to a modern generative video product.
Users submit a text prompt and receive a generated video after an asynchronous processing workflow. The system should support job submission, scheduling, model inference on GPU workers, progress tracking, result storage, and video retrieval.
Start with a high-level architecture, then deep dive into the scheduler. In particular, discuss how the system handles failure cases across the workflow, including scheduler failures, worker failures, retries, duplicate execution, partial outputs, storage failures, and capacity exhaustion.
Quick Answer: This question evaluates proficiency in designing scalable, reliable ML-driven distributed systems, covering asynchronous job orchestration, scheduler design, GPU-based model inference, progress tracking, result storage and retrieval, and fault handling across complex workflows.