Search and ranking over datasets which are constantly evolving in real time is a challenging problem at scale. Updating the documents in the index with real time signals like inventory status and click through rates can improve the search experience considerably. The fields which needs to be updated at scale can be used as hard filters as part of the retrieval strategy or as another ranking signal.
In this talk we’ll present an overview of the real time indexing architecture of Vespa.ai which supports true in-place partial updates of searchable fields, including tensor fields. We also compare the real time indexing architecture of Vespa.ai with search engines built on the Apache Lucene library.