metarank

Why we included this project

Most search and recommendation systems rank by a generic relevance score that ignores what visitors actually click. Metarank is a reranking service you drop in front of your existing Elasticsearch or OpenSearch stack, reordering candidate results with signals like clicks and purchases from real users. You get dozens of prebuilt features such as CTR, user session context, and time-based factors without writing feature extraction code, plus optional LLM-powered bi- and cross-encoder text scoring for semantic reranking. It trains models like LambdaMART out of the box and supports collaborative-filtering style recommendations, serving the trained model over an API for a low-code path to personalized listings and search results. Latency stays in the tens of milliseconds and the service scales horizontally, so it holds up under production traffic rather than just demos.