Large Language Model–Generated Tutoring Responses Receive Higher Ratings but Are Penalized by Source-Based Evaluation Bias
Ola Ozernov-Palchik, Fabio Catania, John D. E. Gabrieli, Satrajit S Ghosh
MGAIC (MIT Generative AI Consortium), 2025–2026
Riverst is a digital-tutor platform for childhood literacy; KIVA is an instance focused on vocabulary learning. Built on Pipecat and senselab, integrating speech-feature extraction. Research questions: Can an avatar understand child speech? Can LLMs adapt to each child? Can an LLM effectively teach vocabulary?

Resources
People
Current
Former
Publications
Ola Ozernov-Palchik, Fabio Catania, John D. E. Gabrieli, Satrajit S Ghosh
Edit this project in src/data/projects.yml. Add a paper insrc/data/publications.yml and tag it with the project slug. SeePROJECTS.mdfor conventions.
Help us count visits
We use Google Analytics to measure page views and improve the site. No advertising, no cross-site tracking. You can change this any time from the footer.