Cave Scan 2026

A technical proof of concept: scan a wine bottle label by photo, extract producer, cuvée, appellation and vintage via Mistral’s vision model, then identify the bottle by matching it against a bundled reference-data sample.

Fields still missing (grape variety, tasting notes, color) are estimated by an LLM and flagged as such; the grape variety’s origin is enriched via Wikidata. Each bottle then joins a personal cellar tracked in a local SQLite database, with a monthly summary and a view that prioritizes bottles by their tasting window.

Freemium offer via RevenueCat (scan quota, cellar size cap on the free tier), GitLab CI with dependency vulnerability scanning, obfuscated release builds.

Technically finished, never published to the app stores: for now, a proof of concept.