Dainius Jocas
Search engineer at Vinted; a recurring author of the Vinted Engineering posts on Vespa — covering recommendation retrieval and low-latency ranking with match-features.
Articles
- Optimizing Vespa Latency with Match-Features at Vinted — match-features for in-engine feature serving + latency cuts
- Adopting Vespa for Recommendation Retrieval at Vinted — two-tower ANN retrieval (with Aleksas Kateiva)
- Dense Retrieval at Vinted — multilingual-CLIP two-tower hybrid semantic retrieval (with Laurynas Jasiukėnas)
- Billion-Scale Search (Search Scaling Chapter 9) — 1B docs on Vespa; summarized in Vinted - Migrating Search from Elasticsearch to Vespa
- Optimize Asymmetric Re-ranking with Algebra — algebraic rewrite of Asymmetric Re-ranking scoring in Vespa, 28% latency cut at 1M documents (personal blog, jocas.lt)