claude code / developer tools / ai trading / buildinpublic
How I Used AI to Code a Feature. 37 Seconds Became 14.3.
I was generating too many events in a minutely backtest on NexusTrade. Claude Code wrote the design. I stashed my changes, profiled the old path, and got 37 seconds on 10 years of history. Then I popped the stash and reran the same script. 14.3 seconds. A 60 percent drop from one change.
The floor with no events at all is about 3 seconds, so there is more to cut. This is the actual loop: profile, let the model propose the change, measure it, keep it if the number moves.
Transcript
0:00This is my github profile, and this is how many contributions I've had in the past year.
0:05I've also used so many AI tokens that OpenAI literally gave me an award for it.
0:11Here's how I use AI to become a more productive software engineer.
0:15For context, I'm building an algorithmic trading platform that operates on minutely data.
0:20Each time step generates events. I'm generating far too many events.
0:23I'm trying to figure out how to reduce the amount of data that I'm generating.
0:26Claude Code generated my final design. Let's see how much faster my backtests are.
0:32The first thing I'm going to do is use git stash to get rid of all of my changes.
0:37Then I'm going to run this profile backtest script to see how long this backtest takes me.
0:42After running the script, I see this backtest takes me 37 seconds on 10 years of historical data.
0:49Then I use git stash pop, and now I'm going to rerun the profiling script.
0:54With this one change, my backtest now takes 14.3 seconds.
1:02A 60% decrease with just this change.
1:07Now granted, it still has a lot to go.
1:09If I don't generate events at all, I know my backtest can be as fast as 3 seconds.
1:15And that's really it. That's how software engineers like me use AI to become better software engineers.
Join the conversation
Loading conversation…