The first default question for any ML trading company: If it works, why are you not trading it yourself?

What statistical tests have you done to back up the claim "the ranking engine has been trained on a large amount of data and has high predictive power"?

I'm having trouble connecting predictive AI models on highly biased portfolios only optimized on historical data to live high risk-adjusted returns and diversified portfolios.

Do you have any numbers for the volatility of your portfolios and/or downside risks?