My main work now is Selflet.ai — AI that turns a body of work into a mind you can talk to: faithful to the source, generative beyond it, tested as conversation, not retrieval. It's the frontier I've been walking toward for thirty years of arriving early to strange places: financing oil and gas in 1990s Russia, running a mining company in the Jordanian desert for the late industrialist Jerry Zucker, building and selling an internet company in New Zealand, joining a quant fund the month the first bitcoin futures began trading. I remain CIO and owner of SC Capital Management, an independent investment firm.
I test theories in public through niche media sites: synthetic biology at Synthetic.com, Cuba at CubaJournal.co, cannabis at High.co, crypto staking at LiquidStaking.com. I've published research arguing that biological motivation works like LLM inference. And I'm perpetually finalizing Earth Tycoon, an autobiographical novel about my career as a banker and executive in the frontier economies of the former Soviet Union and the Middle East — I often find my writing is smarter than me.
MA in developing-country economics. BA in English. I'm also a rum judge.
Patents (Pending):
Systems and Methods for Controlled Conversion of Unsupported High-Divergence Events in Retrieval-Grounded Generative Systems
Systems and Methods for Fidelity-Gated Corpus Manufacturing, Provenance-Anchored Training, and Constrained Agent Deployment
Bidirectional Neural Interface System and Method for Simultaneous Neural Signal Reading and Therapeutic Stimulation
Research interests: "Homeostatic Drive as Policy Precision: Understanding Biological Motivation Through Large Language Model Inference Architecture" This paper proposes that biological motivation works like LLM inference: thirst doesn't encode "drink water"—it narrows your behavioral distribution toward water-seeking, like a system prompt constraining output. Homeostatic equilibrium = high temperature (broad sampling); disruption = low temperature (focused). Reward is a separate training signal, not part of the drive. The framework extends transformer-brain alignment findings from language to embodied cognition and generates testable predictions.