mixedbread Reranking

Searching data using traditional keyword-based search can be challenging and frustrating. We've all experienced situations where the search results were largely irrelevant to our query while we were looking for specific pieces of information. However, mixedbread strives to create a world where search user frustration is a thing of the past. A promising way to boost your user experience is using embeddings-based semantic search systems, which can contextualize the meaning of the user's query, allowing them to return more relevant and accurate results.

Unfortunately, many companies have already built large pipelines and systems around keyword-based search. Migrating to a semantic embedding search would therefore be time-consuming and costly.

Two-stage search flow including rerank

Two-stage search flow including reranking

You can leverage your existing search infrastructure and add a semantic precision boost on top with our reranking models, which perform extremely well on industry-relevant use cases. What's more? They're and perform on par with or even better than many closed-source competitors.

Reranking is applied after a first-stage retrieval step. Keyword-based search systems like Elasticsearch, Solr, or RAG systems can be used to retrieve the top 100+ search result candidates, and our reranking models can be applied in the second stage to get the most relevant results to the top.

Our reranking models are based on the cross-encoder architecture. It's named like that because it encodes both the query and the documents or search results in the context of one another and then calculates similarity values between them.