This project seeks to design a digital learning system in local languages in south Africa. Using machine learning algorithms we want to predict learners literacy levels based on some of their characteristics: sex, languages, school names… The data used to train the models came from a longitudinal study in low-performimg rural primary schools in South Africa. Many Machine Learning Alogrithms (Multinomial logitistic regresssion model, Naive Bayes model, Tree based models, K-Nearest Neighbors model, Extreme gradient boosting model…) have been tested to select the best performing model in terms of accurary.