An adaptive learning platform built for the phone people actually own.
Most EdTech fails in tier-2 and tier-3 India for the same reason: it is built for unlimited bandwidth and English-fluent users on expensive phones. We built for the opposite — patchy 4G, three languages, entry-level Android.
What was breaking
The founder had domain expertise and no engineering team. The audience was 14-to-18-year-olds in tier-2 and tier-3 cities working through competitive-exam content, on entry-level Android phones with intermittent 4G and varying English literacy. The platform had to make rote memorisation engaging enough to sustain a daily habit — without burning through a limited data pack to do it.
The technical bets we made
- 1
Built mobile-first with React Native — single codebase, native performance, sub-50MB install. No web app initially; that was a deliberate decision based on the target user profile.
- 2
Designed an offline-first content sync: 24-hour learning modules cached aggressively, syncs back to the server only on Wi-Fi to save user data.
- 3
Implemented an ML-driven adaptive difficulty engine using TensorFlow — recalibrates question difficulty per learner every 5 questions based on their accuracy, time-to-answer, and concept mastery.
- 4
Localised UI in Hindi, Marathi, and Tamil. Critical: didn't just translate strings — restructured icon-heavy layouts so the app worked even when users couldn't read English headings.
- 5
Built a streak-based gamification system — daily login, weekly challenges, and friend leaderboards through phone-number contact discovery, which is the social graph this audience actually has.
What powers this system
From kickoff to production
Field interviews with 40+ target users in Pune, Nashik, and rural Maharashtra. Surfaced the bandwidth + language constraints early.
Onboarding flow, gamification system, content authoring tools for the founder's content team. Tested with 20 real students before code.
Mobile app + backend + adaptive engine + content pipeline. Beta with 500 users in week 12, full launch week 16.
Iterated on retention drivers — push notification timing, streak mechanics, friend invites. Hit a growing learner base by month 9.
What the architecture changed
The engineering decisions and the business model point the same way. Offline-first sync means a lesson finished on a train still counts, so the daily habit survives bad connectivity. Wi-Fi-only upload keeps the app cheap to run on a limited data pack. And the friend-leaderboard mechanic makes referral a product feature rather than a marketing spend — growth compounds through the contact graph instead of through paid acquisition.
Need something similar?
Book a free 30-min architecture audit. We'll sketch the system you need.
Book a 30-min Architecture Audit