video1.0<iframe src="https://www.loom.com/embed/94f152f64357478097f0a6fcb5d6466c" frameborder="0" width="1864" height="1398" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>13981864Loomhttps://www.loom.com13981864https://cdn.loom.com/sessions/thumbnails/94f152f64357478097f0a6fcb5d6466c-a355b0580684f4ff.gif470.961This Loom explains how the Proplans property search prototype optimizes AI usage and filtering logic. It starts with a Pretor home screen search bar with voice or text input, including 600 ms debouncing and structured JSON extraction so the LLM is not used for raw matching. The prompt returns null for unmentioned fields and applies explicit currency conversion rules to avoid incorrect assumptions during filtering. It then reduces token costs by computing match scores and comparison tables locally, and for the advisor chatbot it sends only the top 5 matched properties while stripping unnecessary data like raw images and internal IDs.