I think the surface of the melt will oxidize so you might need to cover it with inert gas but something like this might be possible. One issue is that the crystals might want to melt if they grow downwards into the molten liquid. I think the temperature gradients might be really important here.
> I think the temperature gradients might be really important here.
Hm, I'm trying to think of ways to maintain an even gradient without unduly stirring up the liquid metal.
Probably sealing the pot-o-bismuth inside another container, where the outer container's fluid is being stirred and circulated to minimize external gradients.
Bismuth crystals can also be grown at low temperature, so maybe they can be shape controlled as well.