HARNESSING LARGE LANGUAGE MODELS AND AI AGENTS FOR CHILD BEHAVIOR ANALYTICS IN DAY CARE: A PROOF OF CONCEPT FOR NEXT-GENERATION PARENTAL INSIGHT USING SIMULATED DATA

Abstract In today’s fast-paced society, parents increasingly struggle to stay apprised of their child’s behavioral and developmental progress during daycare hours. Traditional methods, such as continuous baby monitoring cameras—a trend evidenced by the large number of child monitoring-related apps in the Google Play Store dataset—or regular face-to-face teacher meetings, are often impractical for working families. There is a pressing need for innovative, efficient approaches that offer meaningful insights without excessive parental over-involvement. Recent advances in AI, large language models and computer vision technologies, such as Gemini-2.0 Flash, enable analysis of children’s interactions, emotions, and behaviors within daycare settings. Reflecting findings from the previous studies on “Observations of Children’s Interactions with Teachers, Peers, and Tasks across Preschool Classroom Activity Settings,” which highlights the significance of these daily interactions, it is evident that systematic observation and research on preschool teacher-child interactions are crucial to inform such technology-supported monitoring solutions effectively. This study provides a conceptual understanding and a framework on how Large Language Model Integration with live video feeds allows parents to query about specific events, such as “Did my child participate in group activities today?”—and receive comprehensive, privacy-conscious summaries. This approach bridges the gap between busy lifestyles and meaningful parental engagement, providing reassurance and actionable knowledge while respecting the needs of both parents and children. Thus, hours of camera monitoring are reduced to just a few push notifications or a few chat clicks, which happen in seconds.