Two former Google executives have launched InfiniMind, a Tokyo-based startup tackling the challenge of dark data in corporate video archives. Aza Kai and Hiraku Yanagita, who spent nearly a decade at Google Japan, have developed infrastructure that converts petabytes of unviewed video and audio into structured, queryable business data.

Current solutions force companies to choose between simple object labeling and complex narrative understanding. Earlier approaches could identify objects in individual frames but failed to track narratives, understand causality, or answer sophisticated questions about video content. The breakthrough came between 2021 and 2023 when vision-language models matured enough to move beyond basic object tagging.

InfiniMind recently secured $5.8 million in seed funding led by UTEC and plans to relocate its headquarters to the U.S. while maintaining Japanese operations. The company's first product, TV Pulse, launched in Japan in April 2025, analyzing television content in real time for media and retail companies. Their flagship product, DeepFrame, scheduled for beta release in March and full launch in April 2026, can process 200 hours of footage to pinpoint specific scenes, speakers, or events.