[Your Name] · [Email] · [Phone] · [City, ST]
April 21, 2026
Dear Hiring Manager,
I'm applying for the E5 Software Engineer role on the Instagram Ranking team. Your team's talk at Systems@Scale on the multi-objective value model migration was the first external description I'd seen of the on-platform vs. creator ecosystem trade-off — I've been solving a smaller version of that problem at Pinterest for the last two years and I want to go work on the real one.
At Pinterest I drove the migration of the Homefeed ranking model from a two-tower DNN to a single multi-task transformer serving 540M monthly users. The old system optimized for repin rate; the new one co-optimizes repins, long-click rate, and creator diversity, which is the same frame Meta has publicly described. I shipped the first behind-gate experiment in week 6, not week 20 — by scoping the rewrite to the top 5% of sessions and leaving the fallback path intact. The launch lifted session depth by 4.1%, weekly retention by 1.8%, and ad revenue by $62M annualized. Equally important: the rollout plan let us kill the experiment in 90 seconds if anything looked off, which it did twice in the first month. Both rollbacks were clean.
Before Pinterest I was IC3→IC5 at a Series C startup (Roblox, pre-IPO), where I owned the recommendation stack end-to-end and wrote the first GPU inference serving layer. That range — from 'I'll just deploy it tonight' at a 60-person startup to running a 6-person subteam at Meta's scale — is the muscle I'd bring to Instagram Ranking. The thing that drew me to Meta specifically is that it's one of the few places where a single code change on a Wednesday still reaches a billion people by Friday.
I'd love to talk about the team's current bets on creator-ecosystem signals, and to share the design doc from the Pinterest transformer migration (4 pages, internal-scrubbed) if it's useful for the loop. Happy to do the coding rounds in Python or Hack.
Sincerely,
[Your Name]