Pretty interesting, I recently build an nth order spherical harmonics encoder that can encode the electronic structure of a local environment (of n Å) into a high dimensional fingerprint. We can then use this to search against a big TB dataset of known structures we built to see if we can find analogous configurations. I've started building the structure in the article, I'm interested to see what a search turns up.
Neat! Which spherical harmonics descriptor are you using, and what does your in-house TB dataset cover? What do you plan to do with any matches you find?