Michael Aprossine

Actiquest Lab is developing multimodal AI for autonomous and constantly learning AI sport coaches created on the edge of the Computer Vision (CV) and Generative AI (LLM's) domains. Instead of Video LLM's approach of zero shot scene recognition and reasoning about what happened next on video input we're combining sport-domain-based real-time CV labeling and customized Distillation and Retrieval of Online Corrections (DROC) tools to respond effectively with corrections that generally represent the effective sport coaching process. Actiq AI architecture allows you to enrich CV datasets without streaming video to a central server just using smartphone computing power to collect metadata for dataset labeling. Thus, data augmentation from CV CNN like Yolo with generative LLM works very fast. Auto-labeling of Computer Vision outputs yields 60fps human pose estimation, multiple object tracking and AI decision making at high speed.