We present a novel system that explores how multiple embodied LLM agents in AR (Meta Quest 3) can facilitate contextualized group conversations for second language learners. By capturing the user’s immediate environment, the system generates contextually relevant discussion topics and utilizes GPT-4o for agent responses. While prior work using LLM agents + AR focus mainly on dyadic interactions, we investigate the potential of multi-agent conversations. We share preliminary insights on speaking anxiety, learner autonomy, linguistic risk-taking, and design considerations for future systems.

This project was initially created in Fall 2024 as a final project for COS 436: Human-Computer Interaction, where it earned a peer-voted award for “Best System.” Our paper was later accepted and presented as a Late-Breaking Work at CHI ‘25.

Made in collaboration with Ashley Ponce, Sean Mata, and Aminah Aliu. Advised by Yuhan Liu, Lei Zhang, Amna Liaqat, Varun Nagaraj Rao, and Andrés Monroy-Hernández.