Case Study - English lessons from the game on your screen

Looté is an iOS app for Brazilian gamers learning English. It reads a game screen, explains each line in the context of the scene, and turns what you save into spaced-repetition flashcards.

Client
Looté
Year
Service
iOS Development, Applied AI
Looté’s Capture tab: a fantasy dialogue scene with every recognized sentence outlined

Overview

English from the screen you are already playing

Looté is an iOS app for Brazilian gamers learning English. You photograph or share a game screen, Vision finds its text and groups it into sentences, and a tap on a sentence or a word explains what it means in that scene. It is on the App Store.

Whatever you save becomes a flashcard that keeps the screenshot it came from, and the Review tab schedules those cards with SM-2. The app is written in Portuguese with English and Spanish localizations, and Spanish speakers get their explanations in Spanish.

  • Vision OCR
  • iOS translation
  • On-device model
  • Apple’s cloud
  • SM-2 flashcards
  • Portuguese, English, Spanish

How it works

Each answer shows who wrote it

In Automatic, the default mode, the app picks who answers each job. In the tap sheet and in Scene context, the color of each card names its source, and the onboarding teaches the colors before the first capture. The recording shows a tap on a sentence, Apple’s cloud filling in its card, and a tap on one of its words.
Screen recording: a tap on “I found a weathered map.” opens a sheet, Apple’s cloud fills in a purple explanation card, then a tap on “weathered” opens that word’s explanation

Tap, explanation, word

Green card

iOS translation

With the English and Portuguese languages installed, the Translation framework translates every sentence as soon as a print is captured, so a tap opens on a translation that is already there.

White card

On-device model

Apple’s Foundation Models explain the register, the meaning in the scene, an example and a note. It works without internet but takes a few seconds, and it is what the free plan uses for taps and scene summaries.

Purple card

Apple’s cloud

Private Cloud Compute writes the session recap, because a whole queue of prints doesn’t fit the on-device model’s context. For Pro players it also takes scene summaries and taps. At the daily limit, everything moves to the on-device model on its own.

Share sheet

An inbox for prints

A share extension takes up to 20 images from any app’s share sheet and saves them to an inbox without opening Looté. The app walks them later as a queue, one print at a time.

Review

Spaced repetition

Saving a sentence or a word makes a card with its screenshot. Review schedules the cards with SM-2, and Progress shows the streak, reviews per day, and learned and mature cards.

Subscription

Looté Pro

Pro, sold through RevenueCat, unlocks the queue of two or more prints with its session recap, the Full mode where Apple’s cloud answers everything, and more than 100 saved cards.

Product

From the print to the deck

Captured from the running app, in Portuguese, its source language. One sentence appears twice: once explained by Apple’s cloud, once by the on-device model.
Capturar tab over a sample fantasy dialogue scene, with every recognized sentence outlined and two words highlighted

Capture

Sentence sheet for “I found a weathered map.” with Apple’s cloud card: translation, register, the scene, an example and a cultural note

Apple’s cloud

The same sentence explained by the on-device model, with its translation, register and what it means in the scene

On device

Word sheet for “weathered” with its source sentence and the on-device model’s explanation

Word

Scene context sheet with words you may not know, what is happening in the scene, and a suggestion

Scene context

The third of five prints in a queue: a market stall whose item names, prices and descriptions are outlined

Queue, 3 of 5

Session recap of five prints: what happened across the session, then the important lines with their translations

Session recap

Review card for “I found a weathered map.” with the answer shown above the Again, Hard, Good and Easy buttons

Review

Progress tab with a one-day streak, a reviews-per-day chart, and learned and mature card counts

Progress

Onboarding step that introduces Automatic with the green, purple and white cards

Meet Automatic

Stack & proof

Apple’s frameworks and one dependency

Everything except the subscription comes from Apple. Vision reads the text, Translation and Foundation Models answer, and SwiftData keeps the decks. RevenueCat is the one third-party dependency.
Engines
3

iOS translation, the on-device model and Apple’s cloud

Share sheet
20

Images per share from any app, saved to the inbox

Languages
3

Interface in Portuguese, English and Spanish

Free plan
100

Saved cards before Pro is needed

Technologies

  • Swift
  • SwiftUI
  • SwiftData
  • Vision
  • Translation
  • Foundation Models
  • Private Cloud Compute
  • Share Extension
  • App Groups
  • AVSpeechSynthesizer
  • RevenueCat
  • Swift Testing

Vision’s text recognition reads each print, and a pure, unit-tested grouper turns its lines into sentences. SwiftData stores games, cards and the review log. The share extension shares no code with the app, only an App Group folder. Swift Testing covers the SM-2 scheduler, the grouper and the OCR of every sample scene.

More case studies

A menu scanner that picks what to order, entirely on device

An iOS app that reads a restaurant menu from one photo, checks it against what you logged in Apple Health today and ranks the dishes. The on-device model reads; Swift does the math.

Read more

Your music taste gives you away.

Quem Botou Essa? turns the call’s playlist into a file of suspeitos. A realtime music party game on the web and in a native SwiftUI client, connected to the same room.

Read more

Tell us about your project

Where we are

  • São Paulo
    São Paulo, SP
    Brazil