The Problem
Resellers source faster than they list. Everyone in the trade knows the result by name: the death pile. Great inventory sitting unlisted in a back room, a booth going stale, weak photos, missing measurements, inconsistent pricing, and stock scattered across a store, a storage unit, and a repair pile that nobody can search.
This is not an "AI writes descriptions" product. The job is converting physical inventory into organized digital inventory: listings, ads, curated collections, and eventually local buyer demand. The description is the last step, not the product.
The Capture
A seller opens the app and records one guided video per item, 45 to 90 seconds. The app prompts the walkthrough: front, back, each side. The label, maker's mark, stamp, or signature. Every flaw, by name: scratches, stains, chips, repairs, fading, odors, wobble, wear. Say the dimensions, show the tape measure. Say the material, the era if you know it, the provenance if you have it. Say where it physically lives: booth, shelf, bin, back room. Say the asking price.
One video. Ninety seconds. That is the entire data-entry burden, which matters because the data-entry burden is exactly what created the death pile.
The Extraction
The AI does the clerk work: transcribes the spoken notes, pulls the best frames as product photos, OCRs the labels and serial numbers, suggests category and era (mid-century modern, Art Deco, Hollywood Regency, farmhouse), reads dimensions from speech or the visible tape, builds a structured condition summary from the flaws shown, searches comparable sold listings, and proposes a price range with a confidence level.
The Trust Principle
The part that makes this viable in a trade built on authenticity: the AI does not get to make things up. Every generated fact carries its source tag:
- Confirmed by seller
- Read from label
- Seen in photo or video
- Based on comparable listings
- AI estimate, needs confirmation
- Missing, ask seller
And the high-risk fields (brand, maker, era, authenticity, material, measurements, condition, designer attribution, price) require a human tap to confirm before anything publishes. A vintage seller's whole business is trust; software that hallucinates a Knoll attribution destroys it in one listing. The ledger rule, applied to commerce: every claim shows its receipts.
The Listings
One item, one capture, every storefront. Drafts adapted per platform: Facebook Marketplace simple and local, Craigslist plain and searchable, eBay keyword-rich with item specifics, Etsy tagged by style and era, Chairish polished with provenance, plus Poshmark, Depop, Mercari, a Shopify product page, an Instagram caption with some personality, and newsletter copy. The seller approves; the system fans out.
The Arc
Phase one is the death pile itself: capture to listing. Then booth-level inventory operations (where every item physically lives, what is going stale, what to rotate). Then the demand side: curated collections and local advertising built from inventory that is finally digital. Target users run from single-booth vendors and estate sale operators up through regional thrift chains, and the lab is Denver's vintage and antique-booth scene, which I happen to walk weekly anyway.
Status
Concept, actively being specced. The capture workflow, extraction list, trust principle, and platform matrix are written; inventory operations are mid-draft. The wedge is sharp because the pain is universal and named: nobody in this trade needs the death pile explained to them.