You may use the API to construct AI purposes with out including complexity; there’s no want for vector indexes or a separate question setting. As an alternative, you construct on Microsoft’s personal semantic index, which powers the enterprise search options constructed into the platform. This reduces the workload wanted to construct RAG connections, so you possibly can think about the content material you need to floor in your purposes with out having to spend time fascinated about nearest-neighbor search algorithms.

By constructing on the Microsoft Graph’s SharePoint heritage, you’re capable of shortly floor related content material, lowering the danger of producing hallucinations and errors by focusing operations on particular content material and in particular domains. An AI software for the authorized crew can construct on knowledge within the libraries and lists they use, whereas the gross sales crew will have the ability to work with historic bids and phrases.

Like different Microsoft Graph calls, the retrieval API makes use of HTTP calls with requests embedded within the JSON physique of a POST. This features a question string of as much as 1,500 characters, a listing of SharePoint or connector knowledge sources, in addition to Kusto Question Language (KQL) format filters, and the required variety of outcomes. Filters are a strong software to decide on paperwork with particular metadata, for instance, from a single creator or related to a selected venture or buyer. Filter expressions use KQL and work with any doc property.