Deep dive
The learning loop behind the game
Gamification works only when it serves learning. The core loop is short sessions with immediate feedback, spaced review of items you are about to forget, and visible progress that makes coming back tomorrow feel worthwhile. Streaks, XP and leagues amplify that loop; they cannot replace it.
Design the lesson engine around your subject. Languages need listening, speaking, typing and word order. Exam preparation needs question banks, timed practice and explanations. Coding needs interactive editors. A learning designer working alongside our UI/UX design team defines exercise types and feedback before development starts.
How the lesson engine and progress work
Content is stored as structured data: courses, units, lessons, exercises and the items they practise, such as words or concepts. The app downloads a lesson, runs exercises locally for instant feedback, then sends results to the server. The server updates the learner's mastery of each item and decides what to review next.
Start with a simple spaced repetition rule: items answered wrongly come back sooner, items answered correctly come back later. In the scale tier, a model trained on millions of answers predicts recall for each learner and item. Duolingo has written publicly about its own model, Birdbrain, which estimates how likely a learner is to answer each exercise correctly; you need far less data to get real value from a simpler version.
- Model items, not just lessons, so review can target what each learner gets wrong.
- Keep exercise logic in the app for instant feedback; keep scoring rules versioned on the server.
- Log every answer as an event; it powers analytics, adaptive learning and content fixes.
Adding AI tutoring responsibly
Large language models make two features practical that used to be very expensive: explaining why an answer was wrong in plain language, and conversation practice with a patient partner. Speech recognition with pronunciation scoring makes speaking exercises possible in many languages using cloud services.
AI features need boundaries. Constrain the model to the lesson's level and topic, filter unsafe content, especially for younger learners, and evaluate answers against a test set before each change. Track cost per learner, because heavy users of conversation features can be expensive; caching, smaller models for simple tasks and usage limits on the free tier keep margins healthy. Our generative AI development team builds these with guardrails and evaluation from day one.
Retention, reminders and ethics
Daily reminders, streak protection and leagues increase retention, and they can also become annoying or manipulative if pushed too far. Let learners choose reminder times, respect quiet hours and make it easy to pause. Measure long-term retention and learning outcomes, not just daily opens.
Monetisation interacts with retention in the same way. Limits on the free tier, such as a number of mistakes allowed per day or ads between lessons, should slow learners down without stopping them, because a learner who quits never converts. Test where the paywall appears and what premium includes, and watch completion and thirty-day retention alongside conversion. Regional pricing through both app stores helps in markets where a full-price subscription is out of reach for most learners.
Accessibility matters for learning apps in particular: screen reader support, captions for audio, adjustable speed and options that do not rely on colour alone. Our accessibility testing team checks these against WCAG before launch.
Content, schools and scale
As the catalogue grows, the authoring tool becomes the content team's daily workspace: drafting, reviewing, audio generation, translations and versioning. AI can draft exercises and sentences at scale, but human review remains essential for accuracy and tone. Learner reports on individual exercises, such as a wrong answer being accepted or a missing alternative translation, should flow straight into the authoring tool so the content team can fix them within days.
Schools, coaching institutes and companies are often the steadiest revenue. They need classrooms, assignments, progress reports and admin controls, which turn a consumer app into a B2B product. Plan those in the scale tier, and budget roughly 15-20% of the build cost per year for maintenance, plus speech, AI, hosting and store fees that grow with learners.