Multimodal content embeddings
Design a multimodal content-understanding service that produces embeddings and structured tags for every piece of user-generated content (image + caption, and short video + audio transcript) on a platform ingesting ~3,000 items/sec. Downstream consumers include search, recommendations, and a moderation system, so the embeddings/tags must be consistent and reusable across all of them, produced within minutes of upload, and cheap enough to run on the full firehose. Embedding models get upgraded periodically, and you must handle re-embedding the back-catalog without breaking live consumers.
Clarify scale and constraints first. Propose a clean component breakdown, then go deep on the hard parts (data model, bottlenecks, consistency, failure modes) and name the trade-offs you are making.