I didn’t. I went for an interview and quizzed this multi-million £ business about their architecture: it sounded awful.
I made some diagrams on how I would’ve done it, how it would scale etc and why. They were blown away and wanted me to implement this structure including the job they wanted to hire me for.
They sent a contract over: had the wrong name on it
They corrected the name but I noticed the salary was incorrect
They sent a third and by this time I was offered an interview elsewhere so I went

The hirer then called me to say he was frustrated I hadn’t signed a contract yet making it sound like it was my fault for not wanting to sign an illegitimate contract. he then went on to say that the salary had been reduced, I asked why and they said they felt I wasn’t a senior developer.
So I took the other job and they kept their shitty architecture 💁🏼‍♀️

Oh dear, a scaling problem I solved was replacing some Regex matching with simpler string functions. While I'm a huge fan of Regex, it's unreal how much performance they can suck out of some high-n loops...
I got about 120x out of some critical code thus making a CPU upgrade unnecessary.

I worked once in a company which had this tourist app which should show places on map of the city. Unfortunately it slowed the App down to load more then a couple of places. Their solution was to limit the number of loaded places to teb and prohibited zomming out. I made it handle thousands of places at the same time. Main reason for the Performance issue was, that they sent all data they had about places big, big json objects with large text blobs) to the frontend. This part was easy, I instead sent only the data needed for the map like coordinates and icon type obviously. But still the backend struggeled hard with many objects from the DB, because they built a really shitty orm or what ever this was supposed to be: every line of data retrieved from the DB was immideatly wrapped in some class wich direved from another class which had some magic methods in it which caused some absurd loops over all other obejcts and even more DB queries in unexpected moments and also in the fucking constructor. So it turrned out that the map issue was only the top of the iceberg, since using any data from the DB was extremely expensive. The hard part was to understand the insaness of this abnormination and find the bottlenecks.

I kind of ended up writing my own version of Redis, just an Express server broadcasting events you send it, when I made my first full-stack project.

I rewrote a giant VBA workbook (lots of business formulas, custom pivots of the data) into Java apps/microservices so that new tabs, other reports can be easily added using (JSON) data from the other apps.

In general, I was the only dev in the team that understood that monoliths are hard to change or scale...

I’ve been very unlucky, undertalented, or underfunded with scaling issues. Can’t think of a time I ever successfully scaled a website or application to a faster performance because of any or all of those three factors.