Build a strategy
Start from a hunch, a chart pattern, a paper, or a half-finished script. The agent turns it into precise rules: universe, timeframe, entry, exit, sizing. Pine, Python, or plain English in; an unambiguous, runnable strategy out.
Quantitative research lab
Build the strategy. Test it. Turn it into a scanner.
Hedge AI Labs is a quantitative research lab. Work with our agents to build a trading strategy from an idea or from code you already have, test it with out-of-sample discipline and real costs, and turn the survivors into live market scanners. Then run the whole thing in a simulated account before any capital is at risk.
gap > 0.8 × ATR(20) and open_vol < 0.7 × avg_vol(20) and price > 10 and adv > 500k
| Symbol | Gap | Rel vol | Signal | Sim order |
|---|---|---|---|---|
| MU | −2.4% | 0.58 | Long fade | Filled 118.62 |
| FANG | +1.9% | 0.51 | Short fade | Filled 172.05 |
| SOFI | −3.1% | 0.74 | Vol too high | Skipped |
| CRDO | +2.6% | 0.66 | Short fade | Working 94.30 |
Routed to sim account SA-0417 with slippage and commission modeled
Equity$251,940
Today+$1,940
Open positions3
30-day Sharpe1.31
What we do
Three steps, one pipeline. Start at whichever step matches where your idea is.
Test the strategy
Replicated as written, never tuned. Lookahead-safe, cost-aware, split in-sample and out-of-sample, bootstrap significance, compared to buy and hold. A clear verdict with the numbers and the bugs we found. Negative results published too.
Build a scanner
The rules that passed become a live scanner over the liquid universe of US stocks, ETFs, and futures. Same definition that was tested, so nothing drifts. Hits arrive in the app, by email, by SMS push, or straight into a sim account.
Latest evaluation
Every strategy gets the same pipeline. Negative results are published alongside the positive ones.
Opening gap fade with relative-volume filter
Fade gaps larger than 0.8× 20-day ATR when opening volume is under 0.7× average. SPY, QQQ, IWM. Daily bars, Sep 2023 to Sep 2026. Published 3 Sep 2026.
Return, 3y (SPY)+61.2%buy and hold +58.4%
Max drawdown−10.4%buy and hold −17.2%
Sharpe1.28buy and hold 1.19
Win rate63%94 trades
Profit factor1.47net of 1 bp round trip
Beats buy and hold?Marginally on return, clearly on drawdown. 3 of 3 instruments.
StabilityPositive all 3 years. Bootstrap P(net > 0) 96.4%. OOS Sharpe 1.19 vs IS 1.40.
CaveatRoughly 40% of profit came from April to May 2025. Sample is 94 trades in one dominant regime.
Full report includes per-instrument tables, year-by-year results, exit-rule sensitivity, a cost ladder, and a 10-year extension. Read the full report
Do not fund28 Aug
Dual moving-average crossover with RSI sleeve
SPY, QQQ, ES, NQ. Daily. 12 years.
- Return vs buy and hold
- +94% / +186%
- Max drawdown
- −27.5% / −31.2%
- Sharpe, PF, win rate
- 0.58, 1.14, 68%
Trails buy and hold in every window tested. The regime filter reacts too slowly to cushion a fast drawdown, and the exit rule can strand a position when the regime flips.
Fund with limits21 Aug
Overnight hold with contango filter
ES, NQ. Daily. 5 years.
- Return vs buy and hold
- +71% / +64%
- Max drawdown
- −9.6% / −25.2%
- Sharpe, PF, win rate
- 1.08, 1.38, 58%
Holds across all 5 years. Filter removes 22% of days. Sized for a 2% daily loss cap. Now running in a sim account for 60 days before the verdict is upgraded.
Illustrative only. The evaluations shown are examples of report format and are not results of actual submitted strategies. Not investment advice.
Custom scanners
Any rule you can state precisely, we can run against the live tape. Scanners are built by the same agents that build and test strategies, so a rule that passed testing ships with the same definition that was backtested. No drift between the test and the thing that fires.
# scanner: gap_fade_rvol
universe: us_equities, adv > 500k, price > 10
timeframe: 1min bars, evaluated 09:31 to 10:30 ET
gap: abs(open - prev_close) > 0.8 * atr(20)
volume: open_vol(5min) < 0.7 * avg_open_vol(20)
direction: fade toward prev_close
exclude: earnings within 1 day, halted, index ETFs
deliver: app, sms, sim:SA-0417
sizing: 0.5% risk per hit, stop 0.6 * atr(14)
- DataConsolidated US equity and futures feed, 1-minute and daily bars, session VWAP, options chains for large caps.
- Rule typesPrice and volume, technical indicators, gap and range, relative strength, news and analyst catalysts, earnings calendar, options flow.
- DeliveryIn-app hit list, email digest, SMS push, or straight into a sim account.
- Audit trailEvery hit stores the bar and values that triggered it. You can replay any day and see exactly why a symbol fired or was skipped.
- TurnaroundMost scanners are live the next session. Ones needing new data sources take a few days.
| Time | Symbol | Gap | Rel vol | Signal | Close vs entry |
|---|---|---|---|---|---|
| 09:33 | MU | −2.4% | 0.58 | Long | +1.6% |
| 09:34 | FANG | +1.9% | 0.51 | Short | +0.9% |
| 09:36 | SOFI | −3.1% | 0.74 | Skipped | rel vol over limit |
| 09:41 | CRDO | +2.6% | 0.66 | Short | −0.7% |
| 09:52 | UAL | −1.8% | 0.62 | Long | +1.1% |
| 10:07 | ALAB | +3.4% | 0.69 | Short | +2.2% |
| 10:18 | HOOD | −2.0% | 0.44 | Long | −0.4% |
Illustrative output. Hit rate and outcomes vary by day and regime; the scanner does not know which hits will work, and neither do we until the sim account has enough of them.
Sim trading accounts
Live prices, no real money, fills that cost what real fills cost. An account can be directed by a scanner, by a tested strategy running end to end, or by our research agent trading a mandate you set. Every account keeps a full order log, risk limits, and an auditable track record.
How we test
The same seven steps for every strategy. In order, because each one depends on the last.
Replicate as written
Apply real costs
Split in-sample and out-of-sample
Bootstrap significance
Check by exact timeframe
Compare to buy and hold
Verdict, then sim
No parameter search. No window selection. Negative results published. Strategies that pass go to a sim account before any verdict is upgraded to fund.
Working with the agent
Everything happens in one thread. The agent reads what you give it, asks only what it can't infer, confirms its interpretation, then runs the full pipeline.
H
Describe the idea or paste what you have. Pine, Python, or plain English all work. I'll turn it into exact rules, test it, and build the scanner if it passes.
You
Here's a Pine script. Fades opening gaps on SPY when volume is light.
//@version=6
strategy("Gap fade", overlay=true)
atrLen = input.int(20)
gapMult = input.float(0.8)
…H
Got it. Three things before I run it:
- Instruments: SPY only, or SPY, QQQ, IWM?
- Bar timeframe: daily, or does the gap need intraday bars to fill?
- Should I treat process_orders_on_close as fill-at-close of the signal bar? That's how Pine does it.
You
All three. Daily. Yes. And if it passes, build it as a scanner and route it to my sim account.
H
Running: SPY, QQQ, IWM. Daily. 3- and 10-year windows. 1 bp round trip. Versus buy and hold. No parameters changed. I'll flag anything in the code that doesn't do what it looks like it does. If the verdict is fund or forward log, I'll ship the scanner with the identical rule definition and route hits to SA-0417 at 0.5% risk per hit.
Rules confirmedReplicatedBacktesting 2 of 3BootstrapReportScannerSim routing
Illustrative conversation. Live chat requires a free account.
What we don't do
Optimize parameters, pick windows, or soften a bad result.
Confidentiality
Your code stays private. Publication only with consent, anonymized. Scanners and sim accounts are visible only to you.
Turnaround
Most daily-bar strategies finish in the session. Intraday and options take longer. Scanners are usually live the next session.
Who it's for
Self-directed traders who want an honest answer before risking capital, strategy developers who want independent replication, and investors who want to see how testing is actually done.
Sign up for free today
A free account gets you the agent chat, full evaluation reports, your first scanner, and a sim trading account. No card required.