Yiftach Fabian
Summary
Product manager focused on AI: I design AI products and direct AI coding agents to build them. Twenty years across design, interactive 3D, and product; the last two designing and shipping LLM products end to end, from the PRD and the AI interaction design to a working prototype built with Claude Code and Cursor. I define the data models and API specs, prioritize with RICE, and treat evaluation as a first-class product tool.
Selected work
Self-improving AI agent system for Claude Code. ~98 skills, a 3-phase orchestration pipeline, and a schema-validated state layer driving CLI automation. Sole architect.
Conversational AI murder-mystery with 5 Claude NPCs. Designed the turn-taking interaction and the evidence-state schema; directed the build on Cloudflare Workers with Durable Objects, SSE streaming, and per-session prompt caching.
Product strategy for an AI caregiving venture: market sizing (TAM to SAM to SOM), competitor analysis, personas, go-to-market, and regulatory strategy, packaged as a 20-slide investor deck, then validated with a working prototype.
Stock technical-analysis app. Designed the product, UX, and front end (built with Claude Code): a proprietary momentum indicator, a multi-stock comparison canvas, and a planned AI concierge scoped by a hard non-advisory ruleset.
Experience
- Own the product lifecycle for an AI-enhanced maintenance platform deployed by tier-1 enterprises (Israel Electric Corporation, BAZAN, Israel Railway), driving adoption and ROI on the factory floor.
- Prioritize the roadmap with RICE, driven by customer feedback and product-usage data; translate complex operational problems into precise, testable requirements across web, mobile, and headset.
- Lead and align a cross-functional team of developers, 3D designers, and QA through sprint planning and retrospectives, driving decisions across functions without direct authority.
- Built AI automation (N8N + cloud APIs) that removed repetitive manual steps, and introduced AI-assisted prototyping (Claude Code, Cursor) that turns specs into working demos in hours.