Design a Data Labeling Platform
Medium · System Design · about 45 minutes
Asked at Scale AI, Tesla, OpenAI, Anthropic, Palantir, Anduril.
Design a data labeling platform like Scale AI's. Customers upload images, video, text or LLM conversations with instructions; ~50k concurrent human labelers annotate them (boxes, classes, transcripts, preference rankings); labels are quality-checked and delivered before a deadline. Handle ~10M tasks per day with project-level accuracy targets.