Pick up words as you go.
An iOS app that turns the English you run into in real life — menus, signs, bills, house rules — into a small daily review.

▶ Watch the 1‑minute promo

Project name in Xcode is still SnapLingo; the product name is Magpie.

Textbooks hand you word lists sorted by topic. Real life doesn't. Some days you meet eight words you want to know — on a café blackboard, a bus timetable, a recycling bin — and a week later both the word and where you saw it are gone.

Magpie keeps the two together. You snap a photo, it finds the words worth learning for your level, and it stores them with the photo and the place. Later, the word comes back on a flashcard sitting on top of the same picture.

Point the camera at any English text, or pick a photo from your library. Vision OCR reads every line and records where each word sits in the image.

Claude looks at the recognised text and returns a short list of candidates, each tagged with a CEFR level (A1–C2), plus a scene title such as Bakery counter and one of 12 scene categories (food & drinks, shopping, transport, signage, housing, …). Words are grouped by how useful they are for you:

Untick a recommendation and the app nudges your level estimate down a little; tick a "harder" word and it nudges up. You can also tap any word directly on the photo if the AI missed it.

If you took the photo with the camera, Magpie can remember the neighbourhood (Ponsonby, Fitzroy). It asks once, explains why, and the location never leaves the phone.

The Review tab builds a daily plan: new words you saved recently plus whatever is due. The scheduler is FSRS‑6 (default parameters), targeting 90 % retention; a word with an interval of 21 days or more counts as mastered.

Each card has three buttons — Forgot, Got it, Too easy. The back shows a definition in your native language, an example sentence with translation, a "life tip" where it helps ("½ dozen on a menu means six"), and the exact line from the photo the word came from. Every word page also lists the places you've seen it.

• Photo → words. On‑device OCR with Vision; Claude picks candidates, names the scene, and writes explanations.
• Level‑aware. A 30‑second adaptive "tap the words you know" test sets your CEFR level at onboarding; every save and review keeps adjusting it.
• Daily plan. Separate targets for new words and reviews, a past‑week ring chart, and a month calendar you can tap to replay any day's words.
• Mistakes practice. Words you forgot recently are collected into a quick re‑drill.
• Folders. Your own named folders with icons; a word can live in several without duplicating its review progress.
• Scenes. Every photo is kept as a scene with its title, category, shop name and neighbourhood. Delete a scene and words that only came from it go with it.
• Pronunciation. System speech with a choice of accents, for the word and the example sentence.
• Search, Spotlight and Shortcuts. Inline library search; words are indexed in Spotlight; "Start today's review" and "Snap a scene" are available to Siri and the Shortcuts app.
• Reminders. Optional daily study notification at a time you choose.
• Export. Whole library as CSV (UTF‑8 with BOM) that opens in Excel, Numbers or Anki.
• Themes. Five colour palettes (sky, sand, lilac, sage, blush) for the hand‑drawn home screen.

The interface and all AI‑written explanations follow your native language: English, 简体中文, 繁體中文, 日本語, 한국어, Español, Português (Brasil).

• Nothing is sent to the AI until you say yes. The consent prompt appears after your first photo, not during onboarding, and can be changed any time in Me.
• Only the recognised text goes to Claude. Photos, locations and your library stay on the device.
• Location is requested only for camera shots, only after an in‑app explanation, and is stored locally.
• There is no account and no backend. All data lives in SwiftData on your phone.

SnapLingo/
├── Models/          SwiftData models, CEFRLevel, SceneType, NativeLanguage
├── Services/        OCR, Claude API, FSRS scheduler, DailyPlanner, LevelTest,
│                    PlaceLocator, SpotlightIndexer, AppShortcuts, WordLibrary
├── ViewModels/      AppCoordinator (tabs, scan flow, deep links)
├── Views/           Home · Scan · Review · Study · Words · Folders · Me · Onboarding
├── DesignSystem/    Theme, shared components, empty states
├── Utilities/       Token matching, image helpers, CSV export, demo data
└── Assets.xcassets  Hand‑drawn magpie artwork and scene icons
SnapLingoTests/      Swift Testing suites for scheduling, planning, levels, folders, home, me
Parked/              Features set aside during the 2026‑09 redesign (not compiled)

Requirements: Xcode 26.2 or newer, an iPhone or simulator on iOS 26.2+, and a Claude API key for the AI features.

1. Clone the repo and open SnapLingo.xcodeproj.

2. Create SnapLingo/APIConfig.swift (it is git‑ignored):

   import Foundation
   
   enum APIConfig {
       static let claudeAPIKey = "sk-ant-…"
   }

3. Build and run the SnapLingo scheme.

Without a key the app still scans and lets you tap words on the photo; AI word picking and explanations are disabled.

xcodebuild test -scheme SnapLingo -destination 'platform=iOS Simulator,name=iPhone 17 Pro'

Around 90 Swift Testing cases cover the FSRS scheduler, the daily planner, level estimation, folders, home‑screen copy and the Me stats.

Magpie is an active portfolio project. Done: the core loop, level model, FSRS review, folders, mistakes practice, Spotlight and Shortcuts, seven languages.

Next up:

• Sign in with phone / email / Apple once an account service exists (UI is built, gated behind AccountFeatures.isEnabled)
• iCloud sync via SwiftData + CloudKit
• Bringing back shareable word books from Parked/ once there is a backend to share through

Magpie illustrations are hand‑drawn for this project. Spaced repetition follows the open FSRS algorithm.