The old Yankees

we have submitted :)

Search generate direct pricebooks that user want to find out

Agent Driven Recommendations

DON Helper to help you to proceed with correct pricebooks if they were not clear/correct for your request

Progress diagram

We decided to review users interaction with price books recommendations. Currently, the whole search is based on FTS + semantic mix which lacks context: vectors rely only on item description, UOM, currency, and price. In general, search is missing critical context like category, previous user requests history, additional data we can bring to our search engine while interacting with the user. The goal is bringing our users a useful tool they can use themselves while creating new requests - an agent which asks additional data needed for recommendations, which will adjust to their search requests as used & decrease load from support teams. We're exploring whether to show low-confidence results or keep interacting with the user until we are ready to recommend, letting users refine search queries interactively with an agent. Key questions include which data to embed for the search, how to handle noisy user input, and whether to analyze previous request history for better suggestions. Still no clue how to proceed if there are no strong recommendations exist, we're deciding between fallback options or prompting users for clarification.