Design agentic workflow to generate a 1-hour movie
Company: xAI
Role: Machine Learning Engineer
Category: ML System Design
Difficulty: medium
Interview Round: Technical Screen
## Prompt
You’re asked to design an **agentic workflow** (multiple LLM/tools acting as “agents” under an orchestrator) that can generate a **~60-minute movie** from a short user request (e.g., genre + logline + a few constraints).
## Requirements
- **Input:** a short prompt plus optional constraints (genre, target rating, language, art style, key characters, must-include events, target length ~60 minutes).
- **Output:** a full movie package:
- screenplay / dialogue
- storyboard or shot list
- generated video clips (or scene-level renders), audio (speech, SFX, music)
- final edited timeline (~60 min)
- Maintain **consistency** across the whole film:
- character identity (appearance, voice, personality)
- world facts and timeline continuity
- visual style / cinematography constraints
- plot coherence (no contradictions)
- Maintain good **narrative pacing / rhythm** across 60 minutes.
- The system should support iterative refinement (regenerate a scene without breaking continuity).
## What to cover
Describe:
1. The high-level workflow and the agents involved.
2. What state/memory representations you would use to enforce consistency.
3. How you would manage pacing and hit a target runtime.
4. Tool/model choices at each stage (LLMs, video generation, TTS, editing).
5. Quality evaluation and failure handling (automatic checks + human-in-the-loop).
Quick Answer: This question evaluates a candidate's ability to design agentic, multimodal ML workflows for long-form content generation, covering orchestration of multiple LLMs and tools, state and memory representations for continuity, narrative pacing, and quality/failure handling in an ML System Design context.