Postgres was not good at (for us) - IDF and other corpus based relevancy measures. had to hand roll - thesaurus and missspelling - again possible of course with preprocessing and by adding config files - non Latin alphabet languages. E.g. Arabic - needed filesystem access (we used aws rds so couldn’t do it) to add a dictionary based stemmed/word breaker

We used es or solr for those cases. For English FTS with 100k documents doing it in PG is super easy and one less dependency