Daniel Svonava
Vector Compute @ Superlinked
San Francisco, California, United States
Actions
Daniel is the CEO & co-founder of Superlinked.com, an ML infrastructure platform for building intelligent search, personalization and analytics experiences with vector embeddings that combine structured and unstructured data.
Previously, Daniel was an ML Tech Lead at YouTube, where he built ads forecasting infrastructure that powered the buying flows of $10B/y-worth of ads. Today, Superlinked works with customers with millions of data points and daily queries and Daniel will share learnings from taking these projects from POC to Production.
Your metadata is lonely, MoE embeddings can fix that
Filters, hybrid search, rank fusion, re-ranking.. there is no systematic way to integrate structured data into your vector search. Ever seen a mature elastic search deployment with 100 boosts and custom scoring - where the engineers could still evaluate quality end-to-end and iterate fast? Didn't think so..
To fix this, you need embedding models that understand BOTH your unstructured AND your structured data - the numeric, categorical, relational, spatial, temporal metadata, that is critical for building search, recommendations and agentic retrieval - driving both accuracy for the end-user and results for the business.
In YouTube Ads we solved this problem by training custom data object-level encoders from scratch. But there is an easier way!
We call it the Mixture of Encoders - in this talk I'll introduce the technique, the datasets we open-sourced to help with real-world IR evals and explain how we applied it to drive $10M+ of incremental revenue for a fashion marketplace.
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top