You want to identify actual persons based on one or more profiles (based on e-mail address etc). That is something that is not available out-of-the box in Databricks. The 'unified' in Databricks means you have a single platform for several data topics like engineering, analytics, ML.
What you are looking for is in fact a Customer Data Platform which kan uniquely identify a natural person based on characteristics stored.
How Salesforce does it is probably a secret, but very probably they use a combination of name, address, e-mail address etc to check if there are different profiles pointing to the same natural person.
Basically there are 2 approaches in identifying person: deterministic an probabilistic.
The method I already described is deterministic (based on hard rules). A more advanced technique, which can lead to better or worse results than the deterministic method, is the probabilistic method. Here we try to identify a person based on probabilistic models, so this is a form of statistical learning/machine learning.
A combination of both is also possible.
I already mentioned Databricks itself does not have an own CDP, but that does not mean you cannot build one using databricks.
The methods I described can be applied in notebooks. Also there is an interesting blog on the databricks site about Arc, a probablistic model:
https://www.databricks.com/blog/linking-unlinkables-simple-automated-scalable-data-linking-databrick...
There is a lot to be found online. But beware Salesforce CDP does not only identify persons, there is also the whole UI experience, flexible filtering, creating campaigns etc. That is something I do not see in Databricks.
