Curtis Chen (Chutian)
Curtis Chen (Chutian)
CEO, Quanta Edge · Hong Kong · cchencs@connect.ust.hk
AI is becoming a better trader than most specialists. So the job of people changes: not doing the work, but making the best AI trade effectively.
At Quanta Edge, massively parallel AI agents research, build and trade strategies, from market data to the chips that run them. A handful of people set the goals, the tests the agents must pass, risk and capital.
This resets the race. It is no longer about who has the most talent, but who turns the world’s best AI into trading profit fastest.
What I am building
Quanta Edge is an AI-native proprietary trading firm. Massively parallel AI agents search for strategies, train the models and design the FPGA hardware that executes them; a strategy that used to take a team months takes days. Every candidate must pass the same out-of-sample tests and risk limits before it touches capital, and a person approves every release. We trade our own capital, and we grow by adding markets and computing power, not headcount.
Three bets behind it:
- People make AI trade. The scarce skill is no longer finding the next signal. It is building the system, the tests and the risk limits that make the best models trade well.
- Speed is the advantage. Nobody owns a moat in this reset. Every new frontier model makes our system faster, while large firms first have to reorganise teams built around specialists.
- Model and chip, designed together. Strategies and the hardware that runs them are searched together and judged on one number: profit after costs.
What I have built
Futuristic Group, 2025–2026: the trading firm I founded and ran before this.
Sharpe 11
Prediction markets, on-chain
A high-frequency book on Polymarket’s five-minute BTC markets: +580% in a month, about 3,000 public trades and a top-30 leaderboard finish.
Sharpe 3–4
Crypto alpha, live
Systematic signals on crypto markets, run live through a partner fund across two risk budgets: +6% to +10% net of costs in five months, max drawdown under 1.6% (May to September 2026).
3 DEXs
Designated market maker
Market maker for RWA (real-world assets) on three on-chain perp exchanges, providing primary liquidity for tokenized equities and crypto.
Latest research
EvoQuantLab (arXiv; under review at ICLR 2027). Self-evolving AI agents for quant research: an inner loop searches for signals, an outer loop audits each round for overfitting and rewrites the research rules for the next. Run on the same model and data as four leading AI research agents, it delivered the best out-of-sample return in all three markets tested: China equities, US equities and crypto.
Open questions
- Can a check carry information when a candidate’s evidence is the same size as the noise the search itself produces?
- Where is the line between what the system may rewrite and what it must not, and how do you know, from inside, that it was crossed?
- How fast can a judge be without lying? Replays, simulators and partial builds are cheap and repeatable, and the agent will learn to fool them. What must be measured, and what may be simulated?
Join us
HFT trader
AI researcher
Chief Scientist
If one of the questions above is already yours, especially if you think we have it wrong, write to me.
Earlier: reinforcement learning for high-frequency FX trading at Graphen, under Ching-Yung Lin (former Chief Scientist of IBM); out-of-distribution robustness for computer-vision models at Megvii; recommendation algorithms at Alibaba. HKUST (Ph.D. candidate) · Columbia (M.S. Computer Science) · Peking University (B.S. Economics) · Tsinghua (B.Eng. Engineering Physics). Full CV