Romi Lacatan is our new palmtop, emotional-support AI robot capable of holding a complex, real-time conversation, requiring no turn-taking for speaking. The robot quickly interjects while the user is still talking and generates comments based on what it “sees” and “hears” and on conversation history. We developed a deep-learning model that enables these features, further “humanizing” our original Romi robot that many in Japan use to ease their anxiety and loneliness. Our cloud system coordinates Romi Lacatan’s verbal, physical and visual expressions—including gestures (physical rattles/turns) and “facial” expressions (LCD displays)—to collectively best simulate in-person conversation experiences.