Whenever you're ready, walk me through how you'd approach this.
Design a ranking or recommendation system for Pinterest, including a big-data SQL component and coding for graph traversal and sparse matrix operations.
Clarify the Pinterest surface, users, objectives, data scale, freshness, and latency before proposing an end-to-end design. Cover retrieval, ranking, re-ranking, batch and online serving, offline and online evaluation, and implementation details for graph traversal and sparse matrix operations. Explain how you would detect and mitigate feature drift, training-serving skew, and production failures.