Booking.com's vetting of vector databases for shared GenAI search
Embeddings started as a handful of experiments and became shared infrastructure powering dozens of search and GenAI systems. That forced Booking.com to formally pick a production vector database rather than keep bolting on ad hoc options. The post walks through how they evaluated candidates on real workloads and constraints, not benchmarks alone. For anyone facing the same hoarding choice, it's a rare look at a serious company's selection criteria and trade-offs.