From the chip
to the agent.
I've worked across the whole AI stack: chips at Apple, research at the Technion, agentic products at Salesforce. I take hard problems from a first idea to something people rely on.
Now
I lead evaluation for agentic AI at Salesforce. I find out if an agent works before customers do.
I own evaluation across the entire product, and I'm pushing us to test agents against simulated users. It has already worked in one product. Before that, I led conversational analytics, live with dozens of enterprise customers.
Every layer of the stack.
The parts of AI I've worked on, from silicon to business.
- BusinessImperial College LondonCompleting an MBA, to understand how companies get built around the technology.
- AgentsSalesforceConversational analytics, then evaluation for agentic AI.
- ModelsGloatApplied AI research on transformer ranking models and responsible AI.
- ResearchTechnionM.Sc. in computer science and a first-author CIKM 2022 paper on graph neural networks for molecules.
- SiliconApplePhysical design on the silicon team, where a chip ships exactly once.
Writing
I read new AI research and write about what it actually shows.
Google built an AI scientist. Would NeurIPS take it?
52 domains, and only Python survived
The 0.0005% that makes a model harmful
Podcasts
I host two podcasts, both in Hebrew.
Beyond the day job
- Judging
- Assessing teams and ideas at hackathons, including NASA Space Apps Challenge 2026.
- Mentoring
- Graduate students, through Startup Nation Central.
- Reviewing
- Vetting new AI research for NeurIPS (ethics) and WSDM.
- Speaking
- Public talks for SpaceIL since 2015, and about 600 students taught at the Technion.
Let's talk.
If you're building something challenging, investing in it, or putting a stage together for it, I'd like to hear from you. LinkedIn is the fastest way to reach me.
Message me on LinkedInOr leave a note here. It comes straight to my inbox.