0:47Transcript
I Put $25,000 Into an AI Trading Strategy
Austin Starks puts $25,000 into an AI trading strategy, audits the backtests, and shares his Public Portfolio Challenge. Watch and explore the portfolio.
NEXUSTRADE / VIDEOS
Shorts I filmed while building NexusTrade.
Every video has a transcript.
52 shorts · every one transcribed
0:47Transcript
Austin Starks puts $25,000 into an AI trading strategy, audits the backtests, and shares his Public Portfolio Challenge. Watch and explore the portfolio.
1:41Transcript
A day trading strategy with 423,900 views, coded by an AI agent and backtested on a year of minute data. Tuned and charged zero fees, it still lost 2.40% out of sample while the index gained 1.02%.
0:47Transcript
Does RSI work? Across 222,010 real backtests RSI ranked 23rd of the 34 indicators with at least 500 runs, and strategies using it had a 0.51 median Sharpe against 0.87 without it.
1:22Transcript
One command connected Claude Code to my backtest data. The rule it kept turned $10,000 into $119,942 over 11 years, with a 59% worst drawdown.
2:10Transcript
A 0DTE SPX iron condor creator claimed $30,000 and zero losing days in August. Codified his stated rules and backtested them on minute data: sixteen green sessions, four red, and $840,000 of collateral behind the claim.
1:38Transcript
Backtest a top-five market-cap strategy rebalanced every two weeks: $100,000 grew to $415,000 versus $246,000 for the S&P 500. See the rules, costs and drawdown.
1:14Transcript
Four AI models could not speed up my Rust backtesting engine. GPT-6 Astra named the problem, found a 2014 paper, and cut 470 seconds to 51 warm.
1:26Transcript
My backtests took ten minutes. GPT-6 Astra profiled the Rust engine, cited a 2014 paper, and got the run down to 51 seconds on hardware I already owned.
1:24Transcript
Every rule, backtest, crash and bug behind my $25,000 AI trading account is public. Here is what the live book actually did.
1:39Transcript
The three steps an AI trading agent takes to pick trades: writing a deterministic rule, walk-forward testing it over four folds, and deploying it to paper or live.