Batch LLM Document Enrichment Pipeline With Fleet-Wide 429 Backoff, Dedup, and Replay
Company: Scribd
Role: Software Engineer
Category: System Design
Difficulty: medium
Interview Round: Onsite
Overview: A system design question about a batch pipeline that sends documents to a third-party LLM API and writes enrichment metadata back. It tests fleet-wide handling of HTTP 429 rate limits, atomic work claiming with leases for deduplication, replaying the past five days of data, and observability for backlog, failures, and throttling.
Read the full Scribd Software Engineer interview experience this question came from