Human–AI partnership
A framework for interpreting learning behaviors from classroom recordings in which AI models propose interpretations, experts verify and correct them, and authority, labor, and inspectable evidence are allocated across every stage of the analysis pipeline.
GEM-STEP: multimodal analysis of embodied science learning
A modular framework that combines computer vision, audio processing, and LLM-based late fusion to classify self- and socially-shared regulation behaviors, refined over time with chain-of-thought prompting and active learning from human feedback.
LLMs and embodied computational thinking
AI tools that turn students' embodied exploration into executable code through natural-language-to-code workflows, with prompt engineering and multilingual guardrails for culturally aware, pedagogically aligned interactions. Curriculum co-design and pilots in the U.S. and Israel.