How to build an app like Duolingo: features, tech stack, and cost

Duolingo has 500 million registered users and 37 million people who open it every day. That is not because French is suddenly exciting. It is because Duolingo built a habit loop so well-engineered that missing your streak feels like a real loss.

Most founders building a Duolingo-like app are not trying to out-compete a $6 billion edtech company globally. They are building something more specific: a corporate language training tool, a niche skill-learning product for a vertical they know well, a white-label learning platform for an educational publisher, or a gamified compliance training app for an employer. Duolingo's generic model creates friction for all of them. A custom build removes it.

Here is what that build costs:

The biggest cost variable is content. Building the app is one budget line. Building thousands of exercises, audio recordings, and a content management system for linguists is often a separate and comparable budget line. Most teams underestimate this until they are already in development.

According to Duolingo's 2023 Annual Report, the company's daily active user base grew 65% year over year, with a day-30 retention rate of approximately 20%. That is roughly 3x the industry average for free mobile apps.

How does Duolingo make money, and what are your options?

Duolingo runs a freemium model. The free tier is fully functional. Duolingo Super (formerly Duolingo Plus) costs $6.99 per month or $83.99 per year and removes ads, adds unlimited hearts, and gives access to offline mode. According to Duolingo's 2023 earnings, subscription revenue was $461 million, representing about 80% of total revenue. The rest comes from advertising shown to free-tier users and a separate English certification test (Duolingo English Test) priced at $59 per exam.

For a custom build, your monetization options break down by who you are selling to:

If you are selling direct to consumers, the freemium-to-subscription path is the standard model. Free access with a hearts or lesson-limit system creates natural friction that converts users who want to go faster. A $7-12/month subscription removes it. Expect 3-8% paid conversion on engaged free users as a realistic benchmark.

If you are selling to businesses (corporate language training, compliance learning, onboarding tools), charge per seat: $8-25 per user per month depending on content depth and admin features. A 500-person company paying $15/user/month is $90,000 per year in annual recurring revenue. That is a viable standalone business before you have a single consumer user.

If you are building a white-label platform for an educational publisher or school district, the model is a platform license: a flat annual fee ($30,000-$200,000 per year) plus a setup fee for content customization. This reduces customer acquisition cost and gives you predictable revenue.

Who actually builds a gamified learning app instead of using Duolingo?

Duolingo is free and excellent for general language learning. Most founders who build a custom gamified learning product are not competing with it. They are solving a problem Duolingo's generic model cannot touch.

Corporate L&D teams with compliance requirements. A financial services firm that needs employees to pass a regulatory certification every year cannot use Duolingo. They need gamified learning tied to their exact content, with admin reporting, completion tracking per employee, and integration with their HR system. Duolingo is a consumer app. It has no admin dashboard, no SCORM support, and no reporting API.

Niche language or skill publishers. A publisher with a proprietary Mandarin curriculum for business professionals needs to own the learning experience. Licensing Duolingo-style gamification through a white-label tool costs $50,000-$100,000 per year at any meaningful scale and locks them into someone else's content model. A custom build at $60,000-$110,000 is a better investment in year two.

Edtech startups in non-English markets. Duolingo covers mainstream language pairs well. For a startup building Arabic-to-Swahili or a regional language product for Southeast Asia, Duolingo does not exist as a competitor at all. The gap is the market.

Employers building onboarding products. A franchise operator with 50 locations and 500 new hires per year needs a product-knowledge training tool that looks like their brand, tracks completion by location manager, and feeds into their existing LMS. Duolingo cannot do any of this.

What features does an MVP actually need?

Not all of Duolingo belongs in a first version. These features get a product to market and generate the engagement data you need. Everything else belongs in a later phase.

V1: launch (weeks 1-20, $40,000-$60,000)

User onboarding with language or topic selection, daily goal setting, and social login. The proficiency test option is a nice-to-have; skip it for V1.

A lesson engine with three to four exercise types: multiple choice, fill-in-the-blank, text translation, and word matching. Each exercise type takes 1.5-2 weeks to build end-to-end. Four types built well beats eight types built halfway.

A streak system with daily push notifications. This is the single highest-ROI feature per engineering hour. The streak counter is a simple database field and a background job. The notifications take longer because of platform setup for iOS and Android, but the whole thing is 3-4 weeks including QA.

XP and level progression. A flat 10 XP per lesson with a small bonus for a perfect session is enough for V1. The economy can be refined once you have real engagement data.

Basic progress tracking: lessons completed per skill, overall course map progress, current streak count, and total XP. This feeds both the user profile and the adaptive learning logic in later versions.

Admin content management. You need a way for non-engineers to create, review, and publish exercises. Budget 3-4 weeks for a basic CMS. Without it, every content update requires a developer.

Cross-platform mobile using React Native saves $30,000-$50,000 compared to building native iOS and Android separately. There is no performance reason not to go cross-platform for a lesson-and-quiz app.

V2: growth (add after proving the model, $40,000-$60,000 incremental)

Listening comprehension exercises require audio assets for every sentence. Text-to-speech using a cloud API (Google TTS, Amazon Polly) handles this at low cost for V2. Human-recorded audio is the premium upgrade.

Speaking exercises with microphone scoring add 4-6 weeks and significant per-use API cost. Speech recognition APIs from Google or Azure handle the transcription. Pronunciation scoring on top of that requires either a third-party service (SpeechSuper, Speechace) or a calibrated custom model. Do not build this in V1. It is expensive to build, expensive to run, and high-friction if it does not work well.

Leaderboards and social features. Weekly XP rankings against other learners in your cohort are the social proof loop that keeps competitive learners engaged past the 30-day mark. The leaderboard infrastructure is not complex: sorted sets, a weekly reset job, and a display component. But it requires enough active users to feel real. Below 200 weekly active users per league, leaderboards feel empty.

Offline mode. Pre-downloading lesson content to local device storage matters for commuters and users in spotty-network environments. Add it when retention data shows mobile network issues as a churn reason.

Spaced repetition algorithm. A basic variant of the SM-2 algorithm tracks which vocabulary items a user struggles with and resurfaces them more frequently. This adds a meaningful retention lift at the cost of 3-4 weeks of development. Duolingo's research team published a study in 2016 showing spaced repetition significantly improved word retention compared to massed practice.

V3: scale (only relevant above 10,000 monthly active users)

ML-based adaptive learning predicts optimal exercise difficulty and sequencing per user using performance data. This requires millions of session records to train on. Building it before you have that data is premature. The spaced repetition algorithm from V2 handles most use cases until you reach meaningful scale.

Multi-language audio with human-recorded native speakers. TTS works. Human recordings are better. At scale, the quality difference affects subscription conversion.

A/B testing infrastructure for gamification mechanics. Once you have enough users, testing different XP economies, notification timing, and hearts systems generates real product insights. Before that, it is overhead.

What are the real engineering challenges?

The content pipeline costs more than the app

Every exercise needs a source sentence, a target sentence, audio, difficulty classification, and skill tags. A single language course with 25 skills and 1,000 exercises per skill means 25,000 exercise records. Each audio file needs to be recorded, reviewed, and uploaded.

Duolingo has a full content team and a crowdsourced contributor community. For a custom build, you need a CMS where linguists can create and review exercises and a recording workflow for audio. This pipeline routinely costs as much as the app itself for teams that budget for it properly.

The failure mode we see most often: the engineering team ships a technically complete app in week 20, but there are only 50 exercises in the database. The gamification loop works fine. There is nothing to practice. Users churn in week three and the team spends another two months scrambling to populate content that should have been built in parallel.

Speech recognition takes longer than expected

Getting a user to say "Je voudrais un café" into a microphone and accurately scoring whether their pronunciation was correct is not a solved problem at the API level. Cloud speech-to-text handles transcription well. Scoring how close the pronunciation was to native speech requires an additional layer: either a specialized pronunciation API or calibration work specific to each language and learner accent profile.

Google's Speech-to-Text API supports 125+ languages with word-level confidence scores. The transcription is reliable. The pronunciation scoring on top of it takes 2-3 extra weeks of calibration per language. Teams consistently underestimate this by about 50%.

Budget 4-6 weeks for speaking exercises including integration, testing across device types and accents, and UX for when the microphone does not pick up clearly.

Gamification without content produces a two-week app

According to Sensor Tower's 2023 Mobile App Engagement Report, the average day-30 retention for free mobile apps is under 6%. Duolingo's 20% day-30 retention is the result of daily streak pressure, content deep enough to sustain 30 days of learning, and notification timing calibrated to individual user activity windows.

A custom build with well-built gamification but shallow content will show strong day-7 retention (the novelty of the streak system) and collapse at day-14 to day-21 when users exhaust the available material. The streak system cannot save an app that runs out of exercises.

"The apps that retain users past 30 days have one thing in common: they planned the content library with the same rigor as the engineering backlog," said Luis von Ahn, co-founder and CEO of Duolingo, in a 2022 interview with Lex Fridman. "The technology is the easy part."

Build vs. Duolingo: when does custom win?

Keep using Duolingo or an off-the-shelf LMS when: you are building a structured video course, your audience is mainstream language learners who are already well-served, you have under 500 potential users, or your content fits neatly into Duolingo's existing language pairs and exercise formats. Teachable, Thinkific, and Kajabi handle video-based instruction with basic quizzes. They are adequate for course creators.

Build your own when:

Your content model does not fit a standard LMS. If you need proprietary exercise types (scenario-based decision trees, professional vocabulary for a specific industry, image recognition quizzes), no off-the-shelf platform supports them without expensive workarounds.

You are selling to businesses and need admin reporting. Corporate clients need completion data per employee, integration with their HR system, and custom branding. Duolingo for Business exists but is priced for large enterprise contracts and requires you to use Duolingo's content model.

Gamification is your core differentiator. An off-the-shelf LMS with a "points" feature is not the same as a streak engine, XP economy, hearts system, and league-based leaderboard working together. If retention through daily habit formation is the product, you need to own the mechanics.

You are building a B2B white-label product. If your business model requires selling a branded learning platform to other organizations, you cannot white-label Duolingo. A custom build is the only path.

The payback calculation is straightforward for B2B plays: if you are selling corporate learning at $15/user/month and you have a pipeline of 300 users, the first year generates $54,000 in recurring revenue. The custom platform at $80,000 pays back in under two years before you add a second client.

What does this build actually look like week by week?

Start with the content model (weeks 1-2). How many skills? How many exercises per skill? What exercise types? Map the curriculum before writing code. The database schema depends on this, and changing it after week 6 is expensive.

Then build the lesson engine (weeks 3-4). Get one exercise type working end-to-end: content in the database, user completes a session, earns XP. Everything else builds on this foundation.

From weeks 5-10, add the remaining exercise types, the streak system, push notifications, and XP/level progression. This is the gamification core. Ship to a private beta at the end of this phase and watch day-7 retention numbers.

Weeks 11-16: admin CMS, progress dashboard, polish, and QA. This is also when speaking and listening exercises get built if they are in scope for V1.

After launch, watch lesson completion rates and day-7 retention. A day-7 retention above 25% means the habit loop is forming. Below 15% means something is wrong: content is too hard, onboarding is losing people, or notifications are not landing. Fix the weakest link before adding features.

Building an edtech or gamified learning product? We build custom software for edtech companies and SaaS businesses. See our custom software development services or talk to us about your product scope.