The flexibility is what stands out most. It lets us adjust the architecture down the road, and it also provides a lot of tools that simplify DevOps work. Overall, it saves a significant amount of engineering time.

I get very good support from the Vespa.ai team. They really try to understand our domain and help us develop the first PoC. The speed from getting started to shipping a product to our customers was very fast. Vespa.ai is more than a search backend; it’s a retrieval engine that I can use for complex tasks, like indexing with different models and writing complex ranking profiles, even together with our business. The PoC was easy with Vespa Cloud credits, and moving to enclave solutions was also easy because Vespa.ai provided Terraform and other deployment scripts.

The crown jewel for me is the ranking framework: it offers everything from handwritten expressions to large neural network rerankers running on GPUs. And when you run into special cases, the custom searchers help covering that last mile. I think it is great that you can develop any Vespa application on a laptop, and then scale it to huge workloads as is. Performance is very important for me and the design of Vespa allows to experiment fairly quickly with a bunch of parameters. Vespa is open source and with a coding agent in the repository you can be an expert of the system.