Shelfie -- One-Slide Pitch
Shelfie your shelves. Photograph the wall, get the library.
90-second pitch (read aloud)
You own a thousand books. You can't tell me which ones without squinting at a shelf for ten minutes. That's not a library. That's a collection.
It's 2026. Your phone reads barcodes, sure. It also reads bookshelves now. Every spine, every title, every author, in one photo. Shelfie does the rest.
You walk through the house. You shelfie every wall. The system reads every spine, fetches metadata from OpenLibrary and Google Books, and gives you a real library. Searchable. Filterable. Four views: cover grid, spines on edge, books in stacks, sortable table. Pick the shape that matches the way you think about books.
You search "the big red book about traps" and Shelfie finds Grimtooth's. You ask "what's missing from my Discworld shelf" and Claude reads what you own and recommends three books that fill the gap (and one heretical pick).
The first time a friend mentions a book and you say "I actually own that" with confidence, you realize you never had a library before. Goodreads is a list. LibraryThing is a database. Libib makes you scan barcodes one at a time. Shelfie reads a wall in 30 seconds.
Every executive I know has a wall of books they've never indexed. They're going to want this.
Screenshot mockup
Open mockups/library-home.html in any browser. Five tabs: Library (with four view shapes), Last shelfie (the green / yellow / red review), Gaps, Subjects, Stats. The mockup runs off Edward's two real shelves: a TTRPG cabinet and a fiction shelf. 37 books indexed, 30 confident, 5 verify, 2 questions.
Scope summary
| Stack | Next.js 15 + TypeScript + Tailwind + Claude vision + OpenLibrary + LanceDB |
| Status | Built (2026-04-27). Live and working. |
| Pricing thesis | $9/mo consumer ("Goodreads but real"), $99/yr lifetime, B2B sales to libraries and bookstores |
| Hardest problem | Spine reading at scale, with confidence the user can act on. Half the spines in any photo are tilted, partial, or decorative. The fix is structured tool output that forces a confidence per row, plus OpenLibrary as the truth oracle, plus a green / yellow / red review UI that makes correction take two taps. |
| Differentiation | Other apps: scan a barcode for every book. Shelfie: photograph a wall. Five orders of magnitude faster on a 200-book room. |
What's NOT in scope (v1)
- Multi-user mode (shared libraries between Edward and Dawn)
- Reading progress beyond simple status
- E-book ingestion (this is about physical books on shelves)
- Barcode scanning. Deliberately. The whole point is that you don't need it.
Why now
Vision models finally read tilted spines without inventing titles. OpenLibrary covers most of the long tail. LanceDB makes the local-first version free to run. The reason this didn't exist last year is that the OCR was too brittle. It isn't anymore.
Edward shelfied two of his own shelves while building this. 37 books in two photos. 30 came back fully confident. 5 needed a quick verify. 2 needed a retake. End-to-end took 90 seconds. That's the experience.