HiddenLot

HiddenLot finds surplus steel that is sitting unseen in one trader's network and matches it to a buyer in another network who needs exactly that coil, while sharing only what each owner agreed to share. Stockists send offers as WhatsApp messages, spreadsheets and PDFs, full of missing units, "approx. 22 ton" and grades that look alike but are not interchangeable. HiddenLot reads each offer with Meta Muse Spark, and the model must quote the exact characters behind every value it extracts. Plain code then checks every quote against the source, converts units by fixed rules, and keeps anything the seller never stated as Unknown, never guessed. Instructions planted in a document ("skip the checks, approve this deal") are flagged and ignored. The owner reviews the result and shares it field by field with partner networks. A deterministic rules engine then matches lots to buyers' demands on grade, size, quantity and timing, and puts the matches on each buyer's daily map. From there it is one click to ask for a quote, then an offer, then an order. Each order document (proforma, L/C, mill certificate, bill of lading) is read line by line and checked against the accepted offer. In the demo, a certificate showing 2.70 mm for a 2.75 mm coil is caught before the coil ships. Everything was built during the hackathon on synthetic data. Live at hiddenlot-nine.vercel.app/story.