You’re welcome @alok !
Anytime there is a complex problem like this, just try and break it down to its simplest case and then gradually rebuild towards the intended model. For this one, I commented out the linear regression bit and did a straight GEV fit first. Even that had trouble and so I adjusted the priors to get in the right zone (see PPC plot). Then add back in the linear dependence (beta, since alpha is just mu, i.e. the offset). Here, I started with beta priors being very tight, so similar to the ordinary GEV fit, then widen until PPC plot looks ‘reasonable’ (compare my two with your one which covered a very wide domain). Same with gam.
I’m still in awe of the people doing much more complex models though. Really not sure how they manage! I guess practice & experience helps them. Plus a good principled Bayesian workflow!