I typed one sentence into a trading tool. About a minute later it had written real Python, pulled four years of market data, run a full backtest, and handed me a strategy that lost money over a four year backtest.
Then it did the thing no "AI trading bot" has ever done to me. It told me I was wrong. It explained exactly why my idea was backwards. And instead of just saying so, it ran a second test to prove it and then rebuilt the strategy into something that actually worked.
The tool is Minara’s Strategy Studio. This is the honest walkthrough, red numbers and all. By the end you’ll see why that "Not financial advice" line is the most important sentence in the whole piece.
One sentence in, a quant strategy out
Minara’s pitch is simple: describe a trading idea in plain language and it turns it into a real, backtested quant strategy. You can also feed it a form, a screen-recording of a strategy video, or paste in code. Under the hood it pulls from a library of 500+ factors, writes actual Python, runs the backtest, and shows you the equity curve and the metrics -- then lets you keep refining it by just talking to it.
There are two modes. Single-Asset builds a strategy for one coin. Multi-Asset builds a whole cross-sectional portfolio -- ranking a basket of names on premium factors, going long the best and short the worst. Minara’s own framing: cross-sectional factor models that used to take institutional quant desks years to assemble are now callable in a sentence. That’s the claim. I spent an afternoon stress-testing it.
Strategy Studio: describe an idea, pick Single-Asset or Multi-Asset, and it does the rest.
The warm-up: one coin, one sentence
I started small. I switched to Single-Asset and typed exactly this:
"Design a 15m momentum strategy for my BTC position based on the recent market situation"
That is the entire input. One sentence.
Before writing a line of code it actually read the market: it pulled roughly 200 recent candles, summarized the regime (a sharp sell-off, a partial bounce, high volatility, then a range), called it a "momentum-with-filter" environment, and only then built an EMA-momentum strategy with an RSI and volume filter.
Here’s the backtest it spat out (BTC, 10x, the two months ending 25 Jun 2026):
+3.47% on 4 trades · max drawdown -1.87% · win rate 75% · profit factor 6.84 · Sharpe 3.51.
On its own, +3.47% is nothing to tweet about. But look at the line right next to it: over the same window, BTC fell 21.63%. The strategy didn’t try to hit a home run -- it made a few percent while sidestepping a 21% drawdown. That’s the real product: not "get rich", but "don’t get run over".
Honest caveat: two months and four trades is a tiny sample. A 3.51 Sharpe on four trades is a nice demo of the workflow, not a track record. Onward to the real test.
The main event: a 30-stock portfolio and a faceplant
This is where cross-sectional factors live. I switched to Multi-Asset, picked the TradFi 30 universe (mega-cap tech -- NVDA, MU, MRVL and friends), and asked for a classic institutional setup:
“Rank this universe on a value+quality composite score, go long the top 10 and short the bottom 10, equal-weight, rebalance monthly”
It built the whole thing and ran it over four years (Jun 2022 -> Jun 2026). The result was a bloodbath:
V1: -21.17% total return · CAGR -5.77% · max drawdown -33.83% · Sharpe -0.22 and it badly lagged a plain S&P index.
A textbook value + quality factor portfolio lost money for four years and got crushed by simply buying the index. A normal "AI bot" would quietly hide a result like this behind a cherry-picked demo. Minara put the red numbers right on the screen.
Then it argued with me and it was right
So I pushed back. I told it the strategy lost money and underperformed, and asked straight up: can you actually make this beat a simple index, or is the whole approach just structurally weak here?
Its answer was the moment the tool earned my respect. Paraphrasing: value + quality is structurally weak in this universe over this window. TradFi 30 is mega-cap tech in a momentum-driven bull market. Ranking by "cheap + high quality" systematically shorts the winners and longs the laggards -- exactly backwards. And rather than just assert it, it said it would prove it by inverting the signal.
Not "here’s a fix". First a diagnosis -- then an offer to prove the diagnosis with a test.
The turnaround: +829%, with an honest asterisk
It flipped the signal first, just to prove the point -- then dropped the dead-weight short leg, leaving a long-only momentum+quality strategy. Same universe, same four years:
+829.20% · Sharpe 1.53 · 80% win rate · beats buy-and-hold by 726%. Dropping the short leg alone took Sharpe from 0.17 to 1.53.
And here’s the part I want to underline: unprompted, it flagged the catch. It told me the -42% max drawdown is real and concentrated in three or four names at a time and then offered to fix it. A tool talking you \out\ of over-trusting its own home-run number is rarer than it should be.
The fix: a trend filter, +1047%
I asked how it would handle that drawdown. It added a 200-day trend filter -- only hold names trading above their own 200-day moving average, and step toward cash when the winners start rolling over:
V3: +1047.39% · Sharpe 1.74 · Calmar 2.07 · drawdown ~-40%.
The full arc: a strategy that lost money, the AI’s fix, and the trend-filtered version -- with every drawdown shown. Read those drawdowns. Even the "good" versions sat through 40%+ paper losses. We’ll come back to that.
Not a black box -- you can read the code
None of this is a vibe. It’s real Python on a framework called xstrategy. I clicked the Code tab and read it line by line: a Strategy class, an alpha() method, and factor imports with names like jkp_fcf_me and an operating-earnings-yield z-score for value, plus jkp_qmj_prof (QMJ profitability) and jkp_f_score (Piotroski) for quality -- blended with a NaN-aware combine so names with missing data drop out cleanly.
The actual generated strategy code -- editable, inspectable, and yours to paste your own into.
You don’t have to trust a black box. You can read exactly what it’s doing, change it, or bring your own code.
Trade it
Backtests are theory. The Trade tab is live: perpetuals on Hyperliquid with three modes -- Autopilot (deploy a strategy and let it run), AI Copilot, or Manual. You deposit, pick the strategy, hit Start.
Live perps trading with a deployed strategy. Note the line baked into the product: "Minara may make mistakes. NFA".
A trading product telling you, on its own screen, that it can be wrong -- that’s the right energy. This is real money, real leverage, and a custodial wallet. Treat it accordingly.
Publish it
There’s a public board literally titled "Top strategies, measured honestly". You can publish your own, and others can fork, star, and copy-trade it. Crucially, the leaderboard shows max drawdown right next to the returns -- so a +3,363% strategy also wears its -42% drawdown in plain sight.
The public board -- literally "measured honestly", with max drawdown shown next to every return.
I published my own V3 to that board. Right now it sits at #11 (+1,047% backtest, Sharpe 1.74, -40.53% drawdown), lined up next to creators posting four-figure backtested returns -- every one of them with its drawdown on display.
The cold shower (read this part twice)
Those +829% and +1047% numbers are backtests on a once-in-a-cycle window. 2022–2026 was an AI-megacap melt-up. Momentum strategies print gorgeous backtests inside a trend like that and they reverse hard when it ends.
The 40%+ drawdowns are not hypothetical; they’re sitting in the same results, concentrated in a handful of names. A backtest is not a forecast, and live trading with leverage adds fees, slippage and funding that a backtest understates.
So the honest takeaway isn’t "this prints money". It’s that the tool reasons like a quant: it told me my idea was backwards, proved it, rebuilt it, and refused to hide the risk. That’s the actual product -- judgment on tap, not a slot machine.
What it costs, and how to try it
The free tier gives you 300 credits -- enough to run real backtests like the ones above. Paid plans: Lite $19/mo(1,400 credits), Starter $49/mo (4,000), and Pro $199/mo (20,000 + API access).
Individual pricing. Strategy Studio is free to start.
If you want to poke at it yourself, you can start on the free tier here:
→ minara.ai/app/trade?r=SLASH1S → only 100 free Lite tier plans here
A tool that only ever agrees with you isn’t intelligent -- it’s a mirror. The first time Minara told me I was wrong, it stopped being a toy and started being useful.
