I wanted to do this at one point with cryptocurrencies.

Use historical pricing data, NLP and comment volume on cryptocurrency subreddits and threads, frequency of search terms from Google analytics, major exchange status (Coinbase, Gemini, Kraken, etc. - whether they are up or down as dummy variables), frequency of crypto-related news, etc. all plugged in as features into models in Python and linked with Gemini trading API.

I started working on it and then got lazy (and bad RSI from being at my computer way too much) and it never materialized. Plus, I guarantee somebody else is doing it, or models like that probably aren't too successful anyway.

Could be fun though.