Ensemble retrievers in Data 360
Key Takeaway
An ensemble retriever runs its individual retrievers in parallel, merges their results into one list, reranks that list by relevance to the search request, and passes only the most relevant results to the prompt or agent.
An ensemble retriever (an "ensemble" for short) in Data 360 bundles two or more individual retrievers, so one grounding step can search several sources at once. Use an ensemble when the best answer may come from more than one search or data source.
Search indexes and individual retrievers
Retrievers build on search indexes. A search index is often general, such as an index of an entire knowledge base. An individual retriever applies one search index to a specific use case, through filters and a chosen set of returned fields.
For example, imagine a search index of your entire knowledge base. One individual retriever on that index might return only one category of articles. Another might return only articles from the last 90 days, and a third might return only articles for one geographic region.