SQLite wins in one area that can be crucial -- simplicity. PostgreSQL is a fantastic and robust database, and it's FTS capabilities are almost certainly good-enough (especially if you consider SQLite's to be good-enough), but it is necessarily complex to support the amount of robustness it offers.
SQLite offers a fraction of the robustness of Postgres, but that difference is not necessarily reflected in the correctness/speed of a simple full text search capability meant to support auto-complete (as in the article). Unfortunately I don't have number to back up this assertion, but I'd be very surprised if SQLite FTS was even 10x slower, maybe someone should do that comparison, I'll put it down and make a humble attempt at comparing these two approaches someday.
Good point - but reads never blocked except during a (concurrent) write, which wouldn't likely be much of a problem with the presumably read-heavy autocomplete scenario described here.
Incidentally, even that's not even an issue anymore, if you enable SQLite's write-ahead logging support:
SQLite offers a fraction of the robustness of Postgres, but that difference is not necessarily reflected in the correctness/speed of a simple full text search capability meant to support auto-complete (as in the article). Unfortunately I don't have number to back up this assertion, but I'd be very surprised if SQLite FTS was even 10x slower, maybe someone should do that comparison, I'll put it down and make a humble attempt at comparing these two approaches someday.