My official work was server side verification. Sending codes to people via SMS and Voice. So that's evaluating new SMS/voice providers, checking that the current ones are working, tracking down issues, working with carriers to setup direct SMPP connections for SMS (which thankfully got offloaded to the FB Mobile Partner team), supporting carrier deals. Figuring out how to make it as easy as possible for users to input codes (including working to make the messages readable for platform APIs that allow automatic/assisted code entry)
But even though that's a lot of things and is super important, most days there wasn't urgent work on that, and I'm pretty good at debugging weird stuff, so I debugged a lot of weird stuff. When you have millions (or billions) of users, 'one in a million' stuff happens a lot.
Diagnosing bugs in FreeBSD and fixing them (or convincing upstream to fix them), diagnosing bugs in client platforms and finding workarounds (also trying to get them fixed, but it's hard to get a lot of movement even when bugs are general and just nobody else diagnosed it), diagnosing bugs in ISP networks and finding workarounds (almost nobody follows through on email to whois contacts, even if you're sending from @whatsapp.com), diagnosing issues in our hosting provider's network and convincing them to fix it, etc. If I had stayed longer, I'd have probably started diagnosing bugs in the FB infra and Linux, but I more or less left when we were almost done with our FreeBSD hosts and it took a while to wind down everything, so it was a good transition point to more or less not touch the service where it was running on FB.
In terms of privacy, I'm not going to say much. I think it's fair to say our goal was to only have the necessary private information to provide service. Messages aren't kept on the server beyond delivery, have a limited time frame waiting for delivery, address book is numbers only, no names, etc. Signal certainly stores less information on the server than WhatsApp. You can certainly read about our many failures (no client to server encryption, then poor encryption, and eventually good encryption) among other well reported issues. End to End Encryption took a long time to develop and deploy, but it sets a bar that hopefully other providers will meet (or exceed).