NomadicML, a startup founded by Mustafa Bal and Varun Krishnan, is bringing automation to the overwhelming task of annotating video data for autonomous vehicles. Their platform uses vision language models to turn raw footage into structured datasets, facilitating better fleet monitoring and training.
The company, which recently raised $8.4 million from TQ Ventures and Pear VC, is working with clients like Zoox and Mitsubishi Electric to refine their AI systems by identifying edge cases such as unique driving scenarios involving police officers or specific infrastructure.
Varun argues that Nomadic’s tool goes beyond simple labelling; it acts as an “agentic reasoning system”, understanding the context of actions in video footage. This technology is crucial for developing intelligent machines capable of handling complex, real-world situations without human intervention.
The challenge lies in scaling this process, not just for visual data but also integrating non-visual sensor information from lidar and other sources. Bal notes that the task is “insanely difficult”, but Nomadic’s approach could transform how autonomous systems learn and adapt to their environments.
