Curious about the latest in autonomous systems? In this episode of The Nubius for Startups podcast (special series from the Nebius Robotics and Physical AI Summit and awards competition), host Anubhav Maheshwari speaks with Mustafa Bal, co-founder and CEO of Nomadic ML.
In recent years, the realm of autonomous systems has experienced rapid evolution and innovation, paving the way for groundbreaking changes in how these systems interact with the world around them. At the forefront of this transformative wave is Nomadic ML, co-founded by Mustafa Bal. This pioneering company is setting new standards in the development and deployment of autonomous technologies, and its journey is nothing short of fascinating.
Bal shares his background in ML infrastructure at Snowflake and Microsoft’s AI frameworks team (including work on open-source libraries like DeepSpeed, ONNX Runtime, and PyTorch, and early work around GPT-3 and Azure ML), and explains how Nomadic ML evolved from a general ML tuning solution into a vision AI data layer focused on automating data labeling for autonomous systems.
The conversation centers on identifying long-tail edge cases in operational video and multi-sensor data (dash cam, LiDAR, and logs), using a constellation of open and closed multimodal models, and adding internal innovations to improve contextual motion understanding across video chunks.
They cover the tradeoffs between open vs. closed models (capabilities, inference costs, token pricing), fine-tuning and feedback loops driven by engineers accepting/rejecting/editing surfaced events, and the importance of reaching 95%+ accuracy while minimizing human involvement to optional verification.
Learn how they're transforming the landscape with human-level insights in data processing. Listen now!
Find out more about Mustafa:
Mustafa Bal is the co-founder and CEO of NomadicML (also stylized as Nomadic ML), a San Francisco-based AI company focused on visual AI for physical autonomy. The company specializes in semantic video understanding, turning raw video footage (e.g., from autonomous vehicles, robotics, and fleets) into structured datasets, edge case detection, and actionable insights using generative AI and video-language models to accelerate autonomy development.
Find out more:
LinkedIn: https://www.linkedin.com/in/mstfbl
GitHub: https://github.com/mstfbl
Google Scholar: https://scholar.google.com/citations?hl=en&user=LUPy_-IAAAAJ
Company Website: https://www.nomadicml.com
For more information about Nebius, head to Nebius.com
Follow us on X @nebius.ai