Applied researcher at TopK. Writing on retrieval, recommender systems, and sparse representations.https://martinspisak.com/https://martinspisak.com/blog/freshness-without-compromise/https://martinspisak.com/blog/freshness-without-compromise/Why I'm excited about TopK's semantic_index: most retrieval systems buy quality by trading away freshness. Co-designing the model, inference engine, and database lets you keep both.Wed, 10 Jun 2026 00:00:00 GMTsearchretrievalrecsyshttps://martinspisak.com/blog/late-interaction-has-no-limit/https://martinspisak.com/blog/late-interaction-has-no-limit/The LIMIT benchmark is held up as evidence that multi-vector models struggle. I tried to reproduce it — and found late interaction is embarrassingly good at exactly this task.Fri, 13 Mar 2026 00:00:00 GMTretrievallate-interactionrecsyshttps://martinspisak.com/blog/the-future-is-sparse-compressae/https://martinspisak.com/blog/the-future-is-sparse-compressae/Embedding databases with hundreds of millions of vectors are costly to serve. CompresSAE uses a sparse autoencoder to cut the footprint 10×+ with only a small hit to retrieval quality.Sat, 12 Jul 2025 00:00:00 GMTsparse-embeddingsrecsysvector-databaseshttps://martinspisak.com/blog/instagram-reels-collaborative-filtering/https://martinspisak.com/blog/instagram-reels-collaborative-filtering/Instagram's ranking changes are marketed as a win for small creators. Underneath, they read like a scalability play for early-stage collaborative filtering.Wed, 01 May 2024 00:00:00 GMTrecsysrecommender-systems