Henry Zhang

An LLM proposes trading hypotheses; trusted code decides whether they are true. A bounded research loop built on Karpathy's autoresearch and Loop Engineering: staged holdouts, gate sets, a hash-chained ledger — with a live in-browser demo.

A token price benchmark for LLM inference, and the goods it refuses to index

Run a coding agent like a new hire: a written working agreement, guardrails enforced by hooks, a fresh-context reviewer, plan and decision logs. One-line bootstrap.

A Rust engine for mid-frequency trading: reasoning signals from social and prediction-market data, gated by code. One code path for backtest and live paper, risk that fails closed, deterministic replay, an experiment ledger.

Fast, low-cost real-time agents: take small decisions off the critical path with a System One classifier (Jev). Live demo in your browser.

Plain-English investment observation to a validated US equity screen, ranked names and cited explanations. The model proposes; code decides.