About
Product leader who reads the implementation.
I started in physics and electro-optics, then spent 20+ years turning sensor systems, AI products, and agentic workflows into things people can operate, buy, trust, and scale.
How to read the background
This is not a career-history page for its own sake. The useful fact is that my work keeps returning to the same pattern: when a system is complex enough to fool a dashboard, I go back to the mechanism.
That is the shape of help I sell here. For founders, it becomes senior product leadership without a full-time executive hire. For technical teams, it becomes AI and agentic-system consulting from someone who can evaluate the product, the architecture, and the operational failure modes in the same conversation.
Mechanism before theater
My academic base is unusual for a product leader: dual BSc degrees in Physics and Electrical & Electronics Engineering from Tel Aviv University, taken simultaneously, followed by an MSc in Electrical Engineering focused on AI, machine learning, computer vision, data science, and signal processing.
The instinct became practical early. At TruMedia, a near-infrared people-meter project had stalled after a senior electro-optics consultant modeled eye retroreflection with R3 where the physics required R2. I re-derived the model, proved the geometry problem with my own tests, built the optical bench and synchronized LED/camera setup, and the product received Nielsen Engineering approval before the 2008 crash ended the company.
Then the mechanism became product work
I carried that same habit through Applied Materials in semiconductor mask-inspection optics, Mantis Vision in 3D imaging, Logic Industries and AGT International in large-scale sensing and analytics programs, and BriefCam in enterprise computer vision.
At BriefCam I was Director of Product during my 2018-2022 tenure. I led the product organization, worked inside an AI/ML R&D environment with 30+ PhD algorithm developers, and owned product decisions that helped deployments scale from roughly 40 cameras per installation to 1,200+ cameras per site. I also instituted the AI and computer-vision testability discipline that made algorithm owners define pass/fail criteria before QA handoff.
Earlier, at Logic, I owned sensors and analytics scope inside UAE national smart-city, defense, and critical-infrastructure programs while the project company scaled from zero to roughly 1,000 people. I was later promoted into the CTO office of AGT International.
The range is not a list of interests. It is one unusually wide operating envelope.
Market to product
Market discovery, competitive research, users and pain, product value, packaging, pricing, go to market, and performance marketing.
Mechanism to evidence
AI and machine learning, random forests, physics, electro-optics, imaging, sensors, signals, architecture, QA, evaluation, and acceptance protocols.
Creation to adoption
R&D and delivery, audio, music, and video production, classical and generative methods, product marketing, adoption, cost, scale, and customer satisfaction.
The advantage is not knowing every answer. It is being able to follow one decision across these boundaries, identify the operating envelope, and keep market value connected to what the system can actually deliver.
The AI work is not new to me
My MSc work used graph spectral methods and Israeli Meteorological Service radar data to infer rainfall from microwave attenuation between cellular towers. That was real sensor data, real ground-truth work, and real model evaluation before deep learning became default product vocabulary.
At Sentra, I defined DSPM for AI around data-access governance, shipped Microsoft Purview label-based AI access-control work, joined OWASP's AI Security working group, and co-authored a published framework submitted to EU AI Act drafters. The same access-governance idea shows up in my own agentic systems today: agents get scoped knowledge access, not a blank check.
How I work
Find the real failure mode
Before adding process, I ask what is actually breaking: physics, data, incentives, evaluation, architecture, or ownership.
Define the operating envelope
AI products need explicit boundaries: where they work, where they fail, what evidence counts, and who owns the handoff.
Make quality testable
A model owner who cannot define pass/fail criteria has not finished the product work. QA should not guess what "good" means.
Keep access boundaries explicit
Agentic systems should treat knowledge access as a product and security primitive, not as an afterthought behind prompt guardrails.
Now
Today I am building Guardian Agents, a physical-AI healthcare-operations product focused on operational visibility, quality control, and auditable risk mitigation.
I also run Mindful Designs Haven, a generative-AI print-on-demand studio, and operate my own full-stack agentic AI production environment: 12+ autonomous agents across research, outreach, business development, product operations, analytics, CRM, and DevOps, with a graph-backed knowledge layer, multi-model orchestration, and scoped access boundaries.
Outside the work
I am also a divemaster and swimming instructor. It is not the center of the professional story, but it does fit the pattern: check the environment, respect the limits, debrief what happened, and do not confuse confidence with control.