Recognize
Photograph a label and turn it into structured bottle information, with uncertainty shown clearly.
MVP in development
Capture every dram.
Understand your taste.
A computer-vision-powered tasting journal and personal collection for whisky enthusiasts. Identify a bottle from a phone photo, save the moment, and discover the patterns behind what you enjoy.
GOOD EVENINGYour shelf
YK
RECENTLY ADDED
Japanese Single Malt
Honey · Citrus · Oak
Taste profile12 tastings
One place for every bottle and tasting
Remember
Build tasting notes around your own aromas, flavours, finish, and serving style.
Discover
See how your preferences develop across regions, casks, categories, and time.
How it works
Computer vision finds the label.
Claude understands the bottle.
Whisky labels are difficult in real-world photos: small text, reflections, curved glass, low light, multilingual typography, and visually similar editions. The recognition pipeline combines specialist vision models with Claude's multimodal reasoning.
- 01Recover
Detect the label and improve difficult regions with restoration, super-resolution, and glare-aware preprocessing.
- 02Reason
Use Claude to interpret Japanese and English label evidence and extract name, distillery, age, ABV, category, and cask details.
- 03Verify
Reconcile extracted evidence with a whisky database, return ranked candidates, and make uncertainty visible before the user confirms.
- 04Remember
Turn the confirmed bottle and the user's own aroma, palate, and finish tags into a personal tasting history.
Planned research-led features include multi-bottle shelf capture and using several handheld views to recover label details that are too small or blurred in a single frame.
Building with Claude
From a focused idea to a reliable product.
Claude is the semantic reasoning layer in the recognition pipeline: it interprets multilingual label evidence, produces structured whisky metadata, and explains uncertainty for user correction. Claude Code is used across the Flutter and FastAPI codebase for implementation, tests, evaluation tooling, and iteration by a solo founder.
Founder
Research experience built for a visual problem.
Yuki Kondo Founder · Computer Vision Researcher
Yuki's published work spans image super-resolution, segmentation, small-object detection, and multi-object tracking. His research received the MVA 2021 Best Practical Paper Award. Whisky Snap applies that background to a practical consumer problem: recognizing bottle labels reliably outside a controlled lab.
Whisky Snap is an independent project and is not affiliated with Yuki's employers or academic institutions.
Current build
Recognition PoC
Structured extraction and confidence scoring for Japanese whisky, Scotch, and bourbon bottles.
MVP pipeline
Flutter capture flow, FastAPI recognition service, ranked candidates, and user correction.
Evaluation
A growing test set for label extraction accuracy, candidate ranking, latency, and cost.
Whisky Snapについて
一杯ごとの記憶から、自分の好みが見えてくる。
Whisky Snapは、画像認識によるボトル特定、テイスティングノート、コレクション管理をひとつにまとめ、 ウイスキー愛好家が自分の味覚をより深く理解するためのモバイルアプリです。現在MVPを開発中です。