I build AI systems, not just specs.
Twenty years across design, interactive 3D, and product. For the last two, building production AI products with AI coding agents: I write the PRD, design the product, and direct Claude Code and Cursor to build the working prototype.
AI products I designed and built.
Not screenshots of things other people shipped. These are live systems I architected, coded, and put in front of real users.
MOTY
A system where the AI writes its own skills. ~98 skills, a 3-phase orchestration pipeline with parallel QA, research, and learning agents, and a schema-validated state layer driving CLI automation.
Parley
Cloudflare Workers + Durable Objects, per-session prompt caching, SSE streaming, TypeScript monorepo. I designed the turn-taking model and the evidence-state schema behind every interrogation.
CareAgent
Next.js + Anthropic SDK + Supabase pgvector, with a RAGAS eval harness measuring answers against defined acceptance criteria.
Trade1
React + Vite. Proprietary momentum indicator, comparison canvas, and a planned AI concierge scoped by a hard non-advisory ruleset. New design in build; a live version runs at selectrade1.com.
Enwize / WIZRD
Product lead for a maintenance-ops platform deployed by tier-1 enterprises: IEC, BAZAN, Israel Railway. Roadmap, data-driven requirements, N8N automation.
I am the rare PM who ships the thing. I write the spec, then direct AI coding agents to build the prototype that proves it.
That combination started 20 years ago in interactive 3D and games, moved through product and creative direction for an industrial AI platform used by Israel Electric Corporation, BAZAN, and Israel Railway, and now lives in a daily loop of Claude Code and Cursor. I like problems where the data model is the hard part and the answer has to be correct, not just plausible.
Outside work I build small AI games under the AtomicSpoons name, and a caregiving co-pilot inspired by my mother-in-law, who has cared for her husband through Alzheimer's for over ten years.