News

Who: Jason Mohoney

Where: TMCB 1170

When: December 4th, 11 AM

The Mechanics of Vector Databases

Vector databases are critical infrastructure underlying modern AI systems, enabling applications like semantic search and Retrieval-Augmented Generation (RAG). While the core concept of similarity search is well-understood, building indexes that perform well in practice presents unique challenges. This talk explores the mechanics of vector indexing, moving beyond basic approximate nearest neighbor algorithms to address real-world production needs. We will discuss the complexities of filtered vector search, where metadata constraints often break traditional indexing assumptions, and examine adaptive strategies for dynamic environments where data distributions constantly shift, optimizing index structures on the fly to balance latency and recall. We will conclude by highlighting open problems and future research directions in vector databases and indexing.

Jason Mohoney is a Postdoctoral Associate in the Data Systems Group at MIT, working with Prof. Tim Kraska. He completed his PhD at the University of Wisconsin-Madison under the supervision of Profs. Shivaram Venkataraman and Theodoros Rekatsinas. His doctoral research focused on building I/O-minimizing data systems, with a particular emphasis on efficient vector search indexing, a topic he also worked on during internships at Apple.