Design a Text-to-Video Generation System
Company: OpenAI
Role: Software Engineer
Category: ML System Design
Difficulty: hard
Interview Round: Onsite
Design a Sora-like text-to-video generation platform.
Users submit a text prompt, optional generation settings, and possibly optional conditioning media such as an image. The system generates a short video and returns a downloadable result when the job is complete.
Discuss:
- The user-facing APIs and job lifecycle.
- The high-level service architecture.
- How model inference workers should be scheduled.
- How to handle unstable workers, crashes, retries, and partial failures.
- How to store intermediate and final artifacts.
- How to enforce safety, rate limits, and quotas.
- How to monitor quality, latency, and reliability.
Quick Answer: This question evaluates a candidate's ability to design end-to-end machine learning system architectures for text-to-video generation, covering competencies in model serving and scheduling, distributed systems reliability, artifact storage, and safety and quota enforcement.