By Dave DeFusco

For many students, learning a new language can feel frustrating and impersonal. Most apps rely on memorized vocabulary, repetitive drills and one-size-fits-all lessons that move at the same pace for everyone. But two students in the Katz School’s M.S. in Data Analytics and Visualization are exploring whether artificial intelligence can make learning feel more personal, responsive and human.

Tinotenda Maisiri and Donald Haguma developed Cosynced, a research-driven adaptive learning platform that uses AI to respond to each learner’s strengths, weaknesses and progress in real time. Their first focus is language acquisition, one of the most difficult subjects to teach effectively because it requires listening, pronunciation, memory, rhythm and conversation skills all at once.

“The motivation behind it was language itself,” said Maisiri. “Language is rated as one of the hardest subjects to teach, so we wanted to test our technology to see if it could actually give real-time feedback the way a teacher would.”

Cosynced is still in development, but the first version of the platform was released about a month ago for early testing and feedback. The project combines several AI technologies into a single learning system. A custom retrieval-augmented generation, or RAG, pipeline helps the AI pull accurate contextual information, while a Gemini-powered model has been fine-tuned specifically for language coaching and pronunciation analysis.

The system listens to users speak, evaluates pronunciation and adjusts lessons as learners improve. Maisiri said the goal is to move beyond static lessons and create a learning experience that evolves naturally with the student.

“It’s not scripted, it’s not prerecorded and it’s not just a flash card,” he said. “The more you interact with it, the more it adjusts itself to say maybe this is your weakness or maybe you’re already good at this. That’s the adaptive layer we are trying to bring out.”

One of the platform’s most distinctive features is its real-time speech analysis. During short speaking exercises, the AI evaluates phonemes, fluency and audibility while giving immediate feedback. Learners can practice in quick three-minute sessions designed to fit modern attention spans.

“The trick was to make everything short and precise,” said Haguma. “Attention spans are getting shorter because of social media, so we wanted learners to stay engaged the same way they engage with short-form content online.”

The platform also includes scoring systems, progress tracking, streaks and adaptive lesson generation. If a learner struggles with certain sounds or sentence structures, the AI focuses more heavily on those weaknesses instead of repeating material the student has already mastered.

“What happens is the AI analyzes your scores and patterns over time,” said Maisiri. “Then it decides maybe I should adjust the difficulty or revisit certain areas.”

Cosynced is not just focused on AI tutoring, however; the research team is also studying whether peer-based learning can improve outcomes even more. The platform includes a social feature that matches learners with “study friends” at similar skill levels for asynchronous practice sessions.

Research has suggested peer-based learning can significantly improve retention and performance, and Maisiri believes students often learn concepts more effectively from one another than from formal lectures.

“I discovered myself that I understand things better when a peer explains it,” said Maisiri. “A lot of students told us the same thing. Sometimes another student explains a concept in a way that feels easier to understand.”

Designing that social layer has presented challenges, however. Maisiri said many learners hesitate to admit their actual skill level, especially beginners. “People don’t really want to expose themselves and say they don’t know something,” he said. “We’re trying to create a system that feels comfortable and supportive without making it feel awkward.”

Although Cosynced currently focuses on language learning, Maisiri and Haguma see much broader possibilities for the technology. Their long-term vision is to expand the adaptive teaching framework into STEM subjects such as mathematics, engineering and computer science.

“We hope Cosynced becomes more than just a research project,” said Haguma. “We envision a learning assistant that students can carry in their pocket and use anywhere, whether reviewing concepts before an exam or practicing skills during a commute.”