Building a Jane Street-style quant research terminal with Claude and Jev
Quant research is split into six stages (spread, hedge, test, ship, risk, global flow), using Jev to layer cheap compute under expensive reasoning so Claude is only called when a real decision is needed.
ok this is the f*cking masterclass for building a Jane Street-style quant trading terminal with Claude, JEV engineering included Jane Street has built one of the most sophisticated trading operations on Earth. That's where Horizon fits into the workflow. https://join.horizon.trade/hammertime Billions in trading revenue, thousands of instruments, and an entire research infrastructure designed to capture tiny pricing inefficiencies before anyone else does. I wanted to see how much of that research workflow could be recreated with Claude. Not another dashboard showing BTC prices and RSI indicators, but a proper quant research terminal where every trading idea goes through a structured process before capital gets involved. The terminal is built around six stages: → THE SPREAD: Find pricing discrepancies between related instruments, exchanges, and markets. → THE HEDGE: Calculate exposure, identify offsetting positions, and estimate the cost of neutralizing directional risk. → THE TEST: Run historical simulations, measure drawdowns, and check whether the edge survives fees and slippage. → THE SHIP: Turn validated hypotheses into execution-ready strategies with defined entry and exit rules. → THE GATE: Apply risk limits, position sizing, and validation checks before anything reaches the market. → GLOBAL FLOW: Visualize market activity, liquidity, and opportunities across different regions in one interface. The interesting part is what happens underneath the interface. Claude handles hypothesis generation, coding, and complex research. JEV engineering organizes the workload so cheap calculations handle repetitive checks while expensive reasoning is reserved for decisions that actually need it. A weak hypothesis shouldn't consume the same resources as a strategy that has already survived several rounds of testing. And there's still the historical data problem. Building a beautiful terminal is one thing. Knowing whether the trading ideas inside it have any edge is another. Instead of spending an entire weekend building a backtesting environment, you can describe a strategy, test its historical behavior, and inspect the results. Jane Street spent decades building its research infrastructure. This is a much smaller, experimental version of that philosophy, built with Claude. And the whole thing started with an idea for a trading terminal.
This post quotes another post: x.com/hammertime_one/status/2108916760673964533