Engineering notes
Validation methodology, AI strategy generation, and the architectural choices behind AutoQuant, a local-first algorithmic trading platform.
We gave our charts a copilot. Then we made it impossible for it to lie.
AutoQuant v0.9.12 ships Chart Copilot: ask for support and resistance and the chart draws itself. The LLM never touches a number. This is the architecture, and the two bugs we caught live that prove why prompt discipline is not enough.
Lookahead bias in backtesting: why your strategy lies to you, and how to make it impossible
Lookahead bias is when a backtest uses information that was not available at trade time. It is the most common reason strategies look brilliant in testing and die in live trading. Here is how it sneaks into real code, how to detect it, and how to make it unwritable.
We shipped a daily AI trade ideas feed. Here is how we stopped it from making numbers up.
AutoQuant v0.9.11 publishes swing trade ideas every weekday before the open, with real option strikes picked from the live chain. The hard part was not generating ideas. It was making sure the AI never invents a number. This is the architecture, and the bug that proved it.
One line of pandas quietly changed every backtest. It is why we built a language.
A single line of handwritten pandas in our scanner silently turned enter-once-on-the-signal into enter-every-bar, and it flattered the backtest before we caught it. That bug is why PRIOR exists: a small open-source language for trading strategies where a whole class of backtest lies cannot be written.
We built a real-time market scanner in a weekend. Here is what fired on the S&P 30.
AutoQuant v0.4.0 ships StratScanner, a scanner that watches a universe of stocks, fires the moment a condition is met, and hands you a backtestable strategy on the spot. On Saturday it flagged four names sitting at their lower Bollinger bands. We backtested one of them live.
The backtest killer nobody talks about: human leakage
Every retail trader learns about look-ahead bias, survivorship bias, and overfitting. There's one more bias that contaminates strategies passing every standard robustness check, and it lives in your head, not your code.
Why AutoQuant runs on your machine
AutoQuant is a desktop app, not a cloud platform. Your strategies live on your disk, your backtests run on your CPU, your AI Loop calls hit your API keys directly. Here's what local-first means in practice.
Subscribe to the AutoQuant blog
One post every Friday. Backtest thinking, options mechanics, and AI-assisted strategy discovery. No spam.