Author of this github project mentioned here. I too was fascinated by the potential of triangular arbitrage, and the opportunities do exist (when looking at the raw ask/bid prices), but the volume in the order book is too often too low to cover each step of the opportunity. If orders don't fill quick enough, often the opportunity's already lost within seconds and you're stuck holding leftovers, and that's at the minimum buy you can do on binance.
Tried many different approaches to compensate in a private variation of this repo, though from all the testing we did I'm not confident this could scale to bigger sums reliably, especially if it's likely exchanges are doing this themselves without the limitations of fees or minbuy restrictions. You'll get some profits, yes, but is the amount and success-rate worth it? Questionable. More profitable to invest that time builidng a trading bot in my opinion.