学生用 Claude+Jev 搭交易机器，月入 5000 美元
Claude Opus 5.5 负责设计策略与风控，Jev 以 300-500ms 打分触发交易，覆盖 Hyperliquid 剥头皮、新闻交易与链上调仓，并附回测到实盘的完整流程。
A STUDENT I KNOW BUILT A TRADING MACHINE OUT OF TWO AI MODELS. IT'S CLEARING $5,000 A MONTH. -> Claude Opus 5.5 is the brain. Designs the strategy, runs the backtests, sets the risk rules: stop-losses, position limits, max daily loss. -> A second model, Jev, is the trigger finger. Doesn't write text. Scores probability against fixed rules and fires the trade the second conditions line up. The split exists because Opus takes up to 5 seconds to think. Jev decides in 300-500ms for a fraction of a cent. At HFT speed, that gap is the whole edge. What Jev actually runs: -> Scalping on exchanges like Hyperliquid, reading who's buying, who's selling, where price is heading, every few hundred milliseconds. -> News trading. Reads headlines off Bloomberg and Reuters, tags the asset, scores bullish or bearish, fires before the market catches up. -> Portfolio moves. Tracks smart money and on-chain flows to rebalance positions automatically. The pipeline before real money touches it: -> Hundreds of backtests through Claude Code -> Filtering in TradingView against out-of-sample data -> Paper trading on a fake balance -> Live only once the demo holds up > 99% of strategies lose money. No holy grail here. The edge is infrastructure: fast execution, tight risk limits, testing hundreds of ideas instead of trusting one. Limits that matter: max position size, daily loss cap, a kill switch, and Telegram approval on anything big enough to hurt if the model hallucinates.
这条是引用着另一条写的：x.com/atenov_d/status/2102871540282921262