Our project studies how impressions of a conversation partner change from moment to moment during everyday interactions. Using multi-camera video and audio, we extract second-by-second visual, speech, and language cues and align them with continuous, real-time trait ratings collected during playback. We then build computational models that forecast impression trajectories and pinpoint conversational states where perceptions shift. The work advances basic science of social cognition and informs applied domains such as teamwork, hiring, and cross-cultural communication.