video1.0<iframe src="https://www.loom.com/embed/ffda68912ac94ea299ac7ecce4657165" frameborder="0" width="1670" height="1252" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>12521670Loomhttps://www.loom.com12521670https://cdn.loom.com/sessions/thumbnails/ffda68912ac94ea299ac7ecce4657165-4dd5856b01894f26.gif463.083This Loom explains Emmanuel Lugauch’s challenge architecture for turning provider PDFs into consistent quote data using a two-process pipeline. He describes an ingest process that parses new provider PDFs, normalizes multilingual rate and unit language into a JSON-based provider catalog, and a second process that retrieves data from that catalog using AI without reprocessing the PDF. He notes a pricing decision to use the most expensive price because the system lacks day-level context to determine which price applies. He also discusses building a golden dataset for testing, references OpenAI’s guidance to start with human annotated data, and mentions possible next steps like using a voice agent or email to providers to confirm uncertain prices.