Katalog Kilat

Indonesian social-commerce sellers run their whole business inside WhatsApp, Instagram and Facebook chats. The catalog tools assume clean, one-at-a-time product photos and patient data entry. Real sellers have neither: their products sit in messy group photos on shop counters, and their prices sit in chat captions. So most never build a catalog and answer "how much is this?" by hand, hundreds of times. Katalog Kilat starts from the mess. The seller uploads the multi-product photos they already have, and gets a ready-to-sell Meta catalog back. Muse Spark names the products in each photo, SAM 3.1 segments every instance, and Muse Image cleans each cut-out into a studio shot. A two-stage fidelity gate, run by Muse Spark, rejects any edit that changes the product. A deterministic non-generative fallback is used instead, so no altered product can reach the feed. Muse Spark then writes the bilingual (ID/EN) catalog entry. Prices are never invented. Only the seller's own caption or manual confirmation can set an exported price; AI suggestions appear as grey helper text and never reach any export. Output: a Meta Commerce catalog feed CSV, a photo ZIP, an Instagram caption pack, a WhatsApp price list, and a public share page per product with a WhatsApp order button. The demo video is one real live run on the deployed app: 16 products found from the sample photos, 0 errors.