Please do check up whether there are problems with the leaderboard metrics.

I am also facing the same issues where my local RSME is quite good say at 9.6 but still scores poorly on the leaderboard at say 10.45

Is your local RMSE your validation eval metric? If yes, then you need see if there exists any feature engineering or feature selection done by you, such that it solves for the noise in the dataset. I would suggest examining your features more closely.