Quantization: The Important Bits

In this episode of You Know for Search, host Steve Mazak is joined by Uri Cohen, head of Elastic’s platform product team, to discuss Elastic’s new serverless deployment option. They cover how serverless eliminates the need to manage nodes, sizing, upgrades, or sharding by using a stateless architecture that separates storage from compute. This re-architecture boosts performance—nearly doubling indexing throughput—and introduces features like adjustable “search power” for cost vs. speed tradeoffs, automatic weekly upgrades, and instant access to the latest capabilities like Lucene 10. Designed for simplicity and scalability, serverless is positioned as the preferred way for most users to run Elastic in the cloud.

In this episode, I sit down with David Tippett, a search engineer at Github. We chat about his background and journey into the world of search, focusing on his role at GitHub and how he is improving their search experiences. Our conversation explores the complexities of building effective search, including understanding user intent, the importance of data and indexing, and the challenges of measuring search relevance. We also touch upon the evolving landscape of search with the emergence of AI and multimodal approaches. To close this episode, we talk about the difficulties in justifying investment in search and the critical role it plays across various functions within a platform like GitHub. I hope you enjoy this one as much as I did

Hey everyone, in this episode I speak to Ben Trent, one of Elastics Sr Principal engineers, about Quantization. We recorded this episode a while ago and since then, launched our latest quantization feature, BBQ. So this will be a good primer in prep for leveraging that feature which we touch on briefly at the end. I plan to record another episode covering BBQ specifically so hopefully you stay curious! Show Notes: Fun words - Discretize Centroid - https://en.wikipedia.org/wiki/Centroid K-means - https://en.wikipedia.org/wiki/K-means_clustering NDGC - https://en.wikipedia.org/wiki/Discounted_cumulative_gain#:~:text=NDCG%20is%20often%20used%20to,position%20in%20the%20result%20list. Binning - https://en.wikipedia.org/wiki/Data_binning RabitQ - https://arxiv.org/abs/2405.12497

In this episode, we learn how you can do function calling as part of your RAG application built on Elasticsearch with Ashish Tiwari, our Developer Evangelist in India!

I really enjoyed talking to Mayya Sharipova, an Elastic Engineer who been with the company for 7 years! She has worked on many parts of Lucene and Elasticsearch and in this episode we discuss how our HNSW implementation came to be, how KNN works and when you should use it vs Brute force and more talk about Speed, because it's so important to drive engagement with the apps our community and customers build on top of the Elasticsearch stack. Show Notes: https://www.elastic.co/search-labs/blog/elasticsearch-lucene-vector-database-gains https://www.elastic.co/search-labs/blog/multi-graph-vector-search https://www.elastic.co/search-labs/blog/how-to-deploy-nlp-text-embeddings-and-vector-search https://www.elastic.co/guide/en/elasticsearch/reference/current/tune-knn-search.html

Join me as I speak with Chris Hegarty, our resident Lucene PMC Chair, Java and Search expert at Elastic. You can find Chris on X/Twitter at https://x.com/chegar999. In this episode we discuss many of the latest performance improvements Chris and team have brought to Lucene and Elasticsearch by leveraging newer capabilities in Chips and the JDK itself. You'll learn what SIMD, FFM, FMA and more TLA's mean and why they matter. All in all, I learned a lot in this episode and I hope you find it as informative and fun as I did. Thanks Chris for taking the time to speak with me!! Links from the show: https://www.elastic.co/blog/accelerating-vector-search-simd-instructions https://www.elastic.co/search-labs/blog/vector-similarity-computations-fma-style https://www.elastic.co/search-labs/blog/vector-similarity-computations-ludicrous-speed https://www.elastic.co/search-labs/blog/lucene-and-java-moving-forward-together https://openjdk.org/projects/panama/

In our inagural episode I speak with Serena Chou, our PM for Search at Elastic. We cover our Vector search capability, how we are being leveraged in the RAG workflow, our Roadmap and more. I hope you enjoy this episode and it leaves you excited for more. Upcoming episodes will be with our engineers and architects talking about how we built ES and Lucene, recent advancements in capabilities and performance and much much more.