Part 1 was straight forward - I split the input into overlapping pairs and expanded them according to rules. Obviously the string got large and I noticed I had to expand the same pair multiple times.

For Part 2 I noticed that I could replace the 3 character string of expanded pairs with just two pairs. e.g. CH -> CBH - could be written as CH -> [CB, BH] (note that, new pairs would never be created). So instead I could just keep tabs on the pairs that I observed and update that. I recalled that I might get even more speedup if I would create custom rules for repeated application - that would significantly speed things up - but as code ran fast enough I skipped it.