Co-Founder & CEO @NomadicAI | The Understanding Layer for Physical AI | Prev. @SnowflakeDB @Harvard @TJHSST
San Francisco, CA
Joined April 2011
- Tweets1.8K
- Following140
- Followers170
- Likes1.5K
Mustafa Bal retweeted
"Talk to your customers" is the most repeated advice in startups. @nomadicai cofounder @mstf1bl heard it, agreed with it, and still didn't do enough of it.
"Please don't sit in silos. Please don't sit at your office and just iterate on your product without any feedback whatsoever. Do the absolute bare minimum to be able to talk to your ideal customer."
"You're almost always wrong… when you just go to the person across from you, ask them an honest question… they tell you something that's almost always different from what you thought."
I had a wonderful time chatting with the one & only Schuster Tanger at @TQVentures on our latest at @nomadicai. Thank you for having me! ❤️
A robot picks up a package, moves it, and logs a success. Except the package opened in transit and spilled everything inside.
That's a fail.
Those are the failures @nomadicai finds in millions of hours of footage — and explains why they happened.
In the latest episode of The Study, @mstf1bl takes us through why he and his cofounder set aside the idea they founded the company to build, what a fifteen-board chess simul and a box of baklava did for their sales pipeline, and why founders are almost always wrong about their own product.
Timestamps:
00:27 The problem Nomadic is solving
02:30 Edge cases in autonomous driving
03:44 Why human-first data labeling won’t scale
07:22 The package that opened in transit
08:22 Near-human accuracy at superhuman speeds
12:10 From drawing boxes to reasoning about video
14:04 From 75 interviews to one prompt
18:22 The boom in physical AI data
22:52 Meeting his cofounder at Harvard
24:36 Competition as customer obsession
27:07 Show, don’t tell
33:48 Chess, baklava, and closing deals
36:44 Learning to sell
39:22 Winning over the world’s largest companies
43:05 Why Nomadic is hard to replicate
47:23 Protecting customers’ data
49:48 Advice for founders
1
2
133
Mustafa Bal retweeted
The robot failures that look like successes. A box of baklava. A police car on the shoulder.
The next Study.
Coming soon.
Mustafa Bal retweeted
We are hosting the most brilliant minds in robotics & physical AI with @svrobo ! Join us at our office on Wednesday at 717 Market St where brilliant entrepreneurs and operators like @kevinmpeterson1 (CTO @BedrockRobotics) , Abhishek Bhatia from @ServeRobotics and @Griffin from Zendar will showcase what's their top of mind. Come hungry for engaging content & delicious desserts! 🤖 🧠 🍰
luma.com/6h8hzcez?v=1
For those impacted by Amazon AGI team layoffs, we at @nomadicai are hiring Members of Technical Staff to build visual AI that evaluates the next physical frontiers of AI. Please reach out!
4
114
Mustafa Bal retweeted
Fine-grained subtask annotations improve VLA performance.
We trained a vision + proprioception model that’s now SOTA at generating them, generalizing across embodiments.
93.1 F1@50 on REASSEMBLE, 98.6 on @Amazon Robotics blade insertion. Details below 👇🧵
Mustafa Bal retweeted
1/@Waymo's recent pothole mapping partnership with @waze is a great example of vehicles becoming infrastructure observers.
The next unlock is making this work beyond AV fleets with specialized sensor stacks.
With Nomadic, ordinary dash cam footage can become mapped road events: potholes, hazards, blocked lanes, damaged signs, and more.
Very excited to launch our Nomadic visual data engine for autonomous system developers! Try us out for free now through our website. 🤖🧠🚀
The future of computer vision is agentic.
1/ We built Nomadic around a gap we kept seeing in video understanding: VLMs generate chain-of-thought that's fluent and often correct in structure, but weakly grounded in what's actually in the video.
This limitation shows up in cases like "rolling stops". Is the stop sign actually associated with the vehicle’s lane? Does the ego vehicle ever come to a full stop, or only decelerate?
2
143
Couldn’t agree more! It’s an absolute pleasure to build @NomadicML at the Pear Studio. 🙏
At Pear SF Studio, 216 of the brightest AI founders are building the future—together, under one roof. It’s one of the highest concentrations of AI talent in a single physical space. More than a workspace, it’s a living hub of innovation, serendipity, and unstoppable ambition.
This is where the next generation of iconic companies is being born.🍐@pearvc
Mustafa Bal retweeted
PearX Summer 2024
12/ NomadicML
Enterprise stack for optimizing compound AI systems
As enterprises increasingly adopt compound AI systems to achieve state-of-the-art results, they face exponential complexity in optimizing and maintaining these multi-component architectures. NomadicML enables ML devs to continuously validate and optimize performance across their entire compound AI stack.
Mustafa Bal retweeted
After building heads down with my cofounders the past couple of months, I’m thrilled to finally introduce Animos to the world!
Current generative media is limited — short clips and inconsistencies hamper professional use. Animos breaks these barriers with an agentic approach that transforms simple prompts and multimodal inputs into fully customizable 3D animated videos.
For the first time ever, you can iterate on generated videos with precision —switch out characters, styles, backgrounds, and other details while fully maintaining creative freedom, control, and stylistic integrity— all at a fraction of the cost of current generative video tools.
Experience long-form compositional and consistent video generation like never before.
Join us in redefining animation technology! Email [email protected] to learn more or join our waitlist!
#GenerativeAI #animation #AI #generative
Consistency is so damn important for unlocking enterprise value for generative images.
Enterprises want to be able to templatize what they like. They want to have their own in house models. They want granular control over what they generate.
This requirement is only amplified for video. Is there a world in which we can generate highly consistent videos with studio level customizability and granularity?
Stay tuned.
Mustafa Bal retweeted
One of our key sources of human data is no longer fully “human"!
We estimate that 33-46% of crowd workers on MTurk used large language models (LLMs) in a text production task - which may increase as ChatGPT and the like become more popular and powerful.
arxiv.org/abs/2306.07899
Mustafa Bal retweeted
ChatGPT is cool but I'm most impressed by how OpenAI is able to simultaneously serve a multi-billion parameter model to all the new users trying it out right now (including those coming from front page of HN). Kudos to whoever worked on the inference infra there
Had a great time presenting my internship project from IBM Research at ACM Middleware '18 in Rennes, France! I owe my gratitude to my mentor Shripad and manager @canturkisci for all of their help and support!
1
2
