collect data D from historical api, process it to be useful, train model X with data. D can be any data spaced apart by any type of interval.

X needs to make decisions of buy and selling stocks as accurately as possible. until Day Trading is permited, X must make trades of minimum a day long. X must not risk large amounts. X must give priorities to stocks which have higher success ratios to allocate more money to them.

D must be information dense with respect to timeframe. D must not be redundant for speed.

When I have access to a powerful enough GPU, train X on as many stocks as possible. Then finetune copies of X for multiple stocks. Finally, cash in.