To fulfill these financial calls for, the Net IQ platform doesn’t ship complete paperwork to querying brokers. Entire paperwork can result in costly inference additional down the chain, as LLMs course of outcomes repeatedly to drive the agent workflow. As an alternative, Net IQ buildings the knowledge retrieved from the underlying search engine knowledge, delivering what Microsoft calls “structured proof objects” in addition to passage-level data from unstructured textual content paperwork. This could lead to a a lot larger signal-to-noise ratio than merely querying a search engine, with a deal with delivering data that lets brokers work utilizing fewer tokens.
Utilizing Net IQ in your agent code
The API for Net IQ is a typical REST call, delivering a request object to the Net IQ endpoint. Alongside together with your API authorization key, you’ll ship a question, a set of parameters that management the variety of outcomes returned, the language and area used, and the utmost dimension of the responses and the format used. Responses may be returned in textual content, HTML, or markdown codecs, in addition to extracted passages which can be chosen for context. All different choices return the total doc, so may be costlier to make use of. Markdown is an fascinating various, as it may be used as the idea for giving brokers semantic reminiscences.
Outcomes embody necessary contextual and quotation data, together with net web page titles and URLs, in addition to knowledge about when the positioning was final crawled and the way stale the underlying data is. This can be utilized to enhance grounding and supply extra data that may be included in formatted responses — a lot in the identical means as Bing’s Copilot shows context within the type of footnotes in its responses.