Vespa Why Vespa Vespa was built at Yahoo for search+recommendation at web scale. Its vector search is a side-effect of a broader ranking engine — which is also its advantage. - Hybrid retrieval (BM25 + ANN + ColBERT) composed in one query. - Ranking phases let you run cheap scoring over millions, expensive scoring over tens. - Tensor math (sum, reduce, matmul) in the ranking language — you can express almost any scoring function. - In-cluster embedding inference: encode at write and query time without an external service. - Proven at billions of documents (Yahoo, Spotify, Perplexity). Cost: s…