I build the parts of AI products that decide whether they work.
The parts that never demo well: long-term memory, retrieval that actually retrieves, and interfaces the model generates on the fly. Those are the parts I go after.
Ten years across the stack. Today it is the memory layer inside PAM, a proactive AI manager at Harmix. Before that, quantitative models for a fund near a hundred million in assets, a marketplace platform in Tel Aviv, and, where it all started, point-of-sale systems in Rotterdam. I keep ending up at the seam where research meets production, because that is the one place neither side gets to cheat.
Next stop is a frontier lab. This page is the short version of why.
Endometriosis Screening Model
Advised on and built a model that screens for endometriosis about eight years earlier than a typical diagnosis, a condition that usually takes most of a decade to catch.
Fortune 500 Business OS
A dataset and portal that models how every company in the Fortune 500 actually operates. A SolidJS and FastAPI app sits over 500 firm specs with live financials pulled from an API, Cmd+K spotlight search, per-firm dashboards, same-sector peer benchmarks, and a 106-page print-ready PDF book generated straight from the data.
Document-Editing Agent
An AI agent that edits documents end to end, with auth and a persistent-disk deploy on Render. Multi-stage Docker build, React frontend, Python core.
PAM Memory Pipeline
The retrieval and memory engine behind PAM: isolated per-stage snapshots, failure recovery between stages, and an extract to deep-analysis flow that keeps one clean fact base as the single source of truth. It losslessly compresses 100M tokens of context into 400k, a 250x cut, without losing what the model needs to reason.
Research Engineer
Building the long-term memory layer for PAM, a proactive AI manager. A multi stage pipeline turns raw activity into a fact base the model can reason over, catch conflicts in, and synthesize. On the side I prototype generative interfaces, where the model builds the UI for whatever context it is handed. Applied research that still has to survive production.
Co-Founder
Co-founded a studio that folds AI-assisted coding into real engineering workflows, so a small team ships at large-team bandwidth. Sometimes that means shipping the software, sometimes it means running the workshop. A lean team of five has transformed eight businesses and reached over 500 employees, handing them the macro-context and tools to prosper in the new paradigm. The Fortune 500 Business OS in My recent wins above came straight out of this practice.
Quantitative Developer
Design quantitative models that turn market data into decisions, and decisions into value for shareholders.
Senior Software Engineer
Senior engineer on a mobile creator-video product.
Angel Investor
Backed two creators early and advised on growth. Max Arrow Outdoors hit 100k subscribers in its first two months and grew organically to 1.8M, 1.1M on YouTube, 450k on TikTok, and 275k on Instagram, with viral videos past 100M views and a YouTube best-video feature in 2023. Elijah Windmill built a 30k specialized videography community and landed a partnership deal with DJI.
Co-Founder
Co-founded Rentlyz, a Tel Aviv based rental startup, and helped run it for about a year.
Solutions Architect
Owned the technical vision for a buy-and-sell community marketplace: the architecture, the stack, and the trade-offs that let a small team scale it.
Software Engineer
Built point-of-sale systems at the YouWe studio in Rotterdam: Blue Donut Flowers, a florist POS, and a POS for an Amsterdam university food chain and its events.