Building a one-person AI hedge fund with Jev as the real-time decision engine
A TypeSafe engineer's master prompt is plugged into Claude Code or Codex, with Jev handling real-time decisions on market regime, direction and risk, while position sizing and risk control stay in deterministic code.
A TypeSafe engineer just shared the master prompt for building a one-person AI hedge fund that operates 24/7. I connected it with Jev - paste this into Claude Code or Codex and turn one AI agent into an entire trading operation: > CIO and portfolio manager > quantitative and fundamental research desks > strategy development and backtesting > execution, monitoring and risk control > continuous review while you sleep The architecture uses three layers: Claude or Codex handles deep research, coding and strategy discovery. AgentKit coordinates the specialist agents, workflows, tests and deployments. Jev becomes the real-time decision engine - evaluating market regime, trade direction, setup quality, toxic flow and risk before the system acts. The important part: the AI doesn’t get unlimited control. Deterministic code still owns position sizing, execution limits, maximum drawdown and the kill switch. One person. One prompt. An entire AI trading desk running around the clock. Open the prompt → paste it into your coding agent → connect Jev → build the system. Full prompt below. Enjoy ↓
This post quotes another post: “Jev routes models and skills by compiling a full execution plan per request”