Summary

This article discusses the evolution of HubSpot's AI retrieval infrastructure, focusing on the transition from a proof of concept (POC) to a full-scale production system capable of handling over 20 billion vectors. It highlights the implementation of Vector as a Service (VaaS) and the use of Qdrant as a vector database, detailing the challenges faced and solutions developed, including the migration to a Kubernetes Operator pattern for improved scalability and operational efficiency. The article also emphasizes the importance of automated lifecycle management in maintaining high performance and reliability in a rapidly growing infrastructure.

💭 Your Thoughts

my notes: We isolate clusters by product teams to minimize the blast radius from failures affecting the entire platform. we redefined the contract between VaaS (API layer + data pipelines) and the underlying infrastructure to ensure Qdrant clusters are self-discoverable and do not require any additional code changes when a new cluster is created. abstract cluster metadata into a K8s CRD (Custom Resource Definition)

Data Problems Addressed

Technologies Referenced