Building LexAfriq: Accessible AI for Early Dyslexia Screening in African Classrooms

Introduction

In many parts of Sub-Saharan Africa, children with dyslexia go undiagnosed due to a lack of resources, trained specialists, and culturally relevant tools. LexAfriq was born from a desire to change that — a mobile-first, machine learning-powered app designed to pre-screen children for dyslexia using nothing more than a smartphone.

Why Dyslexia? Why Africa?

While dyslexia is widely studied in Western contexts, early detection tools are almost non-existent in many African classrooms. Teachers may notice a child is struggling, but lack the means to identify why. This gap in accessibility and awareness motivated us to build a tool tailored to the African context — no expensive hardware, no clinical supervision required.

Designing a Multimodal Screening System

LexAfriq uses three main inputs to assess early signs of dyslexia:

• Handwriting analysis: Children trace letters on the screen. A CNN-based model evaluates shape consistency and pressure.
• Eye-tracking: Using the phone's camera and OpenCV+Dlib, we monitor fixations, regressions, and saccades during short reading tasks.
• Gamified cognitive tests: Inspired by the Stroop test, these measure executive function in a child-friendly interface.

All tasks are delivered as part of an interactive experience — children don’t even realize they are being screened. The game-like design ensures natural behavior and more reliable results.

The Technical Stack

• Android (Java) for the mobile app
• TensorFlow Lite for lightweight on-device inference
• OpenCV + Dlib for eye tracking
• Python for training the models
• SQLite for local storage

Results

• 96% accuracy in handwriting classification
• 90.2% sensitivity in eye-tracking module
• 89.2% performance from the game-based assessments

All evaluations were conducted using annotated data and simulated classroom conditions.

What We Learned

• Designing for low-resource environments means building for availability, not just accuracy
• Making the experience invisible to the child improves reliability and removes pressure
• ML models are only as useful as their deployment context allows — mobile UX matters

What's Next

• Preparing the first academic paper submission (currently in draft)
• Planning pilot deployments in Ghanaian schools
• Exploring support for multiple local languages and auditory feedback

Screenshots & Demo

[Insert screenshots of app UI: handwriting task, eye-tracking interface, game screen] > [Optional: embedded video or GIF demo]

Final Thoughts

LexAfriq isn’t just a tool — it’s a step toward equitable AI in education, designed with context in mind. If you're working on low-resource AI, accessible EdTech, or educational research in Africa, I’d love to connect.

Interested in collaborating or learning more? Get in touch