Product
AI iPad App & Web Resources
Children complete assessments and training in the app with additional resources to print extending the experience beyond the screen.
Case study · Education App
Dynamilis
Transforming an EPFL research prototype into an iPad app with 300K+ downloads. Top 5% retention and conversion among 75,000+ apps tracked by RevenueCat.
Co-Founder
& Head of Product
2020
– Present
Overview
From EPFL research to App Store success
Team & context
EPFL Spin-off (QS World #22)
Started as a research project at EPFL, ranked #22 worldwide by QS, to become a startup with a 10-person cross-functional team and school partnerships across five countries.
My Role
Co-Founder & Head of Product
Owned roadmap, experimentation, UX, acquisition (paid & organic), and built the product organization from scratch.
The opportunity
Bringing handwriting assessment from the lab to families
The problem
A hidden crisis in classrooms
Handwriting difficulties affect 10% to 30% of children, but they are often caught too late. Teachers do not have time for individual testing, and parents struggle to measure progress objectively.
The starting point
A prototype trapped in a lab
The core AI technology existed at EPFL, but it had no real users and a tiny dataset. The interface was built for researchers, making it far too complex for families or schools to use on their own.
The challenge
Building a real product
We had to turn complex research into a product that a seven-year-old could use at home, a teacher could fit into a busy classroom, and parents would willingly pay for.
Product loop
A simple three step loop for kids and parents
Children write on the iPad with an Apple Pencil. The app analyzes their movement in real time and automatically opens short games tailored to what they need to practice.
Key decisions
Three choices that made the product work
Accuracy over reach
Native iPad and Apple Pencil
We chose to build specifically for iPad and Apple Pencil rather than a general web app. This narrowed our initial audience, but gave us the high precision needed for real diagnostic quality and trust.
Self-serve practice
Designing for the child first
Instead of a complex dashboard for specialists, we built short, engaging games children could play on their own. This increased weekly practice frequency by three times compared to traditional sessions.
Data advantage
School partnerships for a reliable AI
We offered free pilots to over 50 schools. In return, we collected anonymized writing data from 25,000 children. This massive dataset dramatically improved model accuracy and created a long-term advantage competitors could not match.
Experimentation
App Store conversion: 1.3% to 6.5%
Many parents visited the App Store page but left without downloading. Over six months, we ran structured tests on our visuals and messaging. Shifting from technical features to clear benefits for children increased conversion 5x without changing the core product.
Before (1.3% conversion)
After (6.5% conversion)
Results
The numbers that prove it works
300K+
Downloads
Monthly retention and download to paid conversion ranked in the top 5% among 75,000+ apps tracked by RevenueCat.
+21%
Learning speed
In partner schools, children using the app regularly improved their handwriting speed significantly faster than control groups.
Top 5%
Growth efficiency
App Store funnel efficiency ranked in the top 5% among 405,000+ education apps, achieved on a small budget through organic trust and search optimization.
25K
Unique dataset
Created one of the world's largest structured tablet handwriting datasets, driving continuous research and raising barriers to entry.
Retrospective
What I learned
Simplicity is the hardest product choice
Early lab versions tried to display everything the system detected. Users got overwhelmed. The breakthrough came when we focused on a single promise (test and improve) and hid technical complexity behind simple steps. The hardest choice is deciding what to leave out.
Parents and schools need different promises
Parents want quick reassurance and visible progress. Schools need proof that an app will not disrupt daily classroom routines. Trying to speak to both at once created confusion. Focusing on one clear message first made growth much easier.
Real world data collection is core product work
Building our 25,000 child dataset was not just a side task for engineers. It forced us to design for noisy classroom conditions, established academic trust with teachers, and gave us an advantage that competitors could not easily copy.