Similarity Search

Let's talk about Similarity Search. When we type a question into the Mindbreeze InSpire client, such as "Which deployment options are available for Mindbreeze InSpire?", we see two key areas: a results list and a section containing answers.

Answers

Looking closely at the answers section, we notice that certain text parts are highlighted. These highlight the best possible answers based on our question.

This is precisely what Similarity Search does. It scans the results to find text parts fitting as close as possible to the original question. The highlighted text represents the most relevant section. Additionally, we get a bit of context before and after, building a complete and meaningful answer.

Configurable

Of course, we can configure how many sentences should be included in the main answer—the highlighted text. This flexibility is essential because these answers are out-of-the-box processed by the Mindbreeze InSpire Retrieval Augmented Generation (RAG) whenever a question is asked.

Similarity Search & AI Chat

When we open the AI chat side-by-side with the Similarity Search and ask the same question again, we notice that the generated answer originates from the same sources as the top three results of the Similarity Search and contains similar content.

This happens because Similarity Search serves as the foundation for retrieval augmented generation.

The first three answers from Similarity Search are used in the RAG pipeline, where they are processed, and a cohesive, well-structured response is generated and returned to the user.