Spaced repetition is the best-validated vocabulary acquisition technique in the learning science literature. Most language apps implement it badly: fixed intervals that don't adapt to the learner, word pools that don't account for semantic relationships, and review sessions that don't connect new vocabulary to previously learned words that share morphological roots or semantic fields.
Our platform has 4.2 million language learners across 28 language pairs, and we're rebuilding the vocabulary acquisition model from first principles. You'll design models that estimate each learner's memory decay rate per word (faster for visually similar words, slower for words with L1 cognates), optimize review schedules to maximize long-term retention per minute of study time, and build the semantic clustering that introduces new vocabulary in conceptually related groups.
The learning science team — two researchers with cognitive science PhDs — will challenge every modeling decision with domain expertise. You need to be comfortable defending your architectural choices with both empirical evidence and theoretical grounding.