advanced practice · system design prompt
Search platform end-to-end
Design a product search platform: ingestion → analyzer → inverted index → query parser → ranking → result page. Identify where you would add a learned ranker.
The mechanism
A search platform splits into an ingestion/indexing path (durable, batchy, write-optimized) and a query path (low-latency, cacheable, read-optimized). Segment files, an object-storage tier, and an eval harness tie them together.
Expected evidence
Pipeline diagram + the index update strategy (full rebuild vs near-real-time).
Try this drill in the practice workspace →
Hints when you need them
- Recall first, precision second — bad recall is unrecoverable downstream.
- Two-stage ranking: cheap recall, expensive rerank.
Watch for these mistakes
- Designing the query path before the indexing path
- No segment/compaction story for updates
- Forgetting evals as part of the architecture