From Legacy Search to Vespa: What a Real PoC Taught Us
summary
Germany’s largest classifieds platform runs 60 million items with ~1M new items/edits/deletes per day and moves the Motors homepage feed from Elastic to Vespa in a staged PoC: 4 weeks of development, legacy and Vespa running in parallel, AB tests and guardrails. Vespa features used include weighted set, document processors, embeddings stored for approximate nearest neighbor search with GPU preprocessing. Motors sees clear lift with weighted sets; Fashion needs careful normalization and split queries; Home & Garden uses embeddings but triggers higher latency so retrieval runs in parallel and undergoes offline evaluation.
AI-written summary · about