I've been considering something similar - how has that worked out for you? Or, are you not trading and just gathering historical/live data to see whether there might be a repeatable trading signal?
Using just 3 days of data, after the over/under price got to 75% agreement, the ending price for that 15-minute window closed in the same direction a little over 80% of the time, subtracting Kalshi fees your winnings would have been 1-2% at most.
The next script I have in my mind is to find markets that are the same on Kalshi and Polymarket and write a script that interrogates both markets to see if an arbitrage wager exist, a wager where you can bet both sides of the market and still win and would it be profitable enough after fees to make it interesting.
One of the models I’ve been working on is for upcoming Week 1 NFL games.
Using an API I had Claude AI write a script using the API to pull the point spread, over/under and money line current odds from 10 different sports books and update a spreadsheet, then update the spreadsheet at least 3 times per week to detect the trend of how money was being wagered.
This one took about 30 minutes to get working, I had to change a couple of settings within Claude AI which were suggested by Claude AI to make it all work.
Finally, I’ve only begun to work on this project in Claude AI, I’m thinking about starting a “faceless” YouTube channel , I don’t want to film myself reading a script so working with Claude AI to write the script I found a tool that integrates with Claude AI that allows you to record 2-3 minutes of your voice reading anything and using that recording it can clone your voice for the 15 minute script that Claude AI generated along with all the graphics I need to upload to YouTude for a working video.