Skip to content
JL/
Menu

AI Technical & Product Strategy ·Team Building & Leadership ·Physical AI

Jonathan Lwowski

AI/ML Leader • Product, Platforms & Physical AI

I’m a technical and product AI leader who builds teams and puts complex AI into production. From PhD research in control systems and robotics engineering, I went on to build Zeitview’s 13-person AI department and now serve as Field CTO for Physical AI & Robotics at Nebius. At Zeitview, I set product strategy and led AI and autonomy delivery with full-stack product partners, improving analysis quality by 29% and cutting operating costs by 27%. At Nebius, I work with a six-person team and customers on Physical AI Workbench and serverless simulation and training proofs of concept. I connect strategy, technical direction, coaching, and field delivery.

Start a conversation →Connect on LinkedIn (opens in a new tab)
Jonathan Lwowski

Currently building Physical AI at Nebius.

A few things I'm proud of

0 → 13
AI organization built and led
27%
operating cost reduction at Zeitview
200k+
daily requests on a platform I co-founded

About

My work began in control systems research and hands-on robotics engineering, then grew into leading AI teams and products. At Zeitview, I built a 13-person AI department spanning data labeling, ML science, ML engineering, and autonomy. We worked with full-stack product teams for each inspection vertical, improving analysis quality by 29% while cutting operating costs by 27%. Today, as Field CTO at Nebius, I work with robotics customers and a six-person Physical AI team on the infrastructure behind the Physical AI flywheel, including Physical AI Workbench and serverless simulation and training proofs of concept.

I like connecting customer problems to product strategy and technical direction, then staying close to delivery. Building a team is part of that work: hiring thoughtfully, coaching engineers and team leads, setting clear priorities, and making space to learn from what happens in the field.

  • AI & Robotics Strategy
  • Product & Roadmap Ownership
  • Building & Running AI Teams
  • Computer Vision & Perception
  • Physical AI Workflows
  • ML Platform & MLOps
  • Customer Discovery
  • Working with Leaders & Stakeholders

Leadership

My career has moved from control systems research and robotics engineering into building AI organizations and shaping Physical AI products. I connect product strategy and technical direction with delivery and coaching. At Nebius, I help robotics teams put data collection, training, simulation, evaluation, and field feedback into a working Physical AI flywheel, including work on Physical AI Workbench and serverless simulation and training proofs of concept.

I do my best work in senior leadership roles where a company needs to connect an AI strategy to a real product, grow a strong technical team, and build the data and evaluation foundations that real-world systems need.

A few examples

A few public examples of the team, product, and technical work I have led.

  1. Case study 01

    Building Physical AI infrastructure with NVIDIA

    Nebius | Field CTO, Physical AI & Robotics

    Challenge
    Give robotics teams one workflow for data collection, simulation, and evaluation instead of a pile of disconnected tools.
    Leadership
    Co-lead the Physical AI team, contributed to Physical AI Workbench and serverless simulation and training proofs of concept, and co-led a joint demo with NVIDIA and Black Forest Labs at GTC Berlin.
    Scale
    27+ structured customer interviews shaping three solutions built on Isaac Sim, Isaac Lab, GR00T, and Cosmos.
    Outcome
    An open-source workflow that connects data, simulation, and evaluation into one learning loop for robotics teams.

  2. Case study 02

    Building an applied AI organization

    Zeitview | Director of AI/ML

    Challenge
    Establish an AI organization and operating model that could support multiple product verticals.
    Leadership
    Built a 13-person department, hired AI and data-labeling leads, and set hiring priorities, product and technical strategy, roadmap, and team budget.
    Scale
    A 13-person department spanning data labeling, ML science, ML engineering, and autonomy, working with full-stack teams across five inspection verticals.
    Evaluation
    For a deployed high-severity defect alerting system, evaluated missed defects, false alerts, human review, and field tests against release targets of human-level detection and fewer than one false positive per mission.
    Outcome
    Improved analysis quality by 29%, helping attract customers and grow revenue, while cutting operational costs by 27%. Wind and solar analysis reached roughly 90% full automation, cutting their costs by over 95% and customer turnaround from about two weeks to roughly one hour.

  3. Case study 03

    Commissioning vision-guided robotic picking

    Plus One Robotics | Senior Machine Learning Engineer

    Challenge
    Bring computer vision into robotic picking cells that had to work at customer sites.
    Leadership
    Owned integration end to end, including software setup, tuning, and delivery testing; shared final sign-off with project partners.
    Scale
    Commissioned picking cells for four major logistics clients.
    Outcome
    Increased picks per minute by 18% to meet a customer target, helping win contracts against competing systems.

  4. Case study 04

    Scaling a real-time analytics product

    GammaEdge | Co-Founder

    Challenge
    Create and operate an analytics platform that customers could use at meaningful daily volume.
    Leadership
    Co-founded the platform and grew the product alongside its technical and operational foundations.
    Scale
    More than 200,000 daily requests.
    Outcome
    Serves more than 200,000 daily requests.
Read more about how I lead →

Insights

View all

Short guides that expand on the leadership and product choices above.

  1. Go/No-Go Decisions for Machine Learning Products

    Read the guide →

  2. Plan the Path From Prototype to Product

    Read the guide →

  3. Why Physical AI Teams Become Infrastructure Companies

    Read the guide →

  4. Evaluate the Decision, Not Just the Model

    Read the guide →

  5. Define the Problem Before the Model

    Read the guide →

AI & PM Insights on LinkedIn (opens in a new tab)

How I work

For me, building a useful ML product is less about rushing to train a model and more about making good decisions together. I like to start with the real problem, be honest about the constraints, and give the team a clear next step.

Read the AI & PM Insights guide that inspired this framework (opens in a new tab)

  1. 01

    Start with a problem people actually feel

    I begin with the customer decision and the outcome that needs to improve. When the value, user, or measure of success is still fuzzy, it is usually time to learn more instead of building a model.

  2. 02

    Be honest about the data

    I look closely at what data is available, how representative it is, and what it will take to label, govern, and maintain it. Those realities should shape the plan from day one.

  3. 03

    Agree on what good looks like

    Before optimizing anything, I want the team to agree on what good looks like. A useful evaluation connects model behavior to a real decision, workflow, or customer outcome.

  4. 04

    Keep the first version small and useful

    Use the simplest model and integration that can test the product bet. It is much easier to plan for ownership, human intervention, monitoring, and iteration while the work is still small enough to change.

  5. 05

    Decide what to do next together

    Bring product, engineering, data, and operations into the conversation. The answer might be invest, narrow the scope, learn more, or choose a simpler solution. A clear answer helps everyone move forward.