Staff Engineer, Platform
Distributed systems | Cut p99 latency 8x on a 40M-user platform
I work on the boring infrastructure that makes product teams fast. Nine years of backend and platform work in Python, Go and a lot of Postgres. My favourite projects are the ones where the diff is small and the graph moves a lot. I have run on-call for systems doing 30k rps and rebuilt two of them without a maintenance window.
p99 latency reduction
8x
feed service, 40M users
Infra cost saved
41%
feed read path
Tech lead, 6 engineers
Owner
2023-02 — 2024-06
Tech lead, 6 engineers
What did you do, and why that approach?
Replaced a monolithic feed service with a sharded read path and a precomputed ranking cache, migrated live across 40M users with no maintenance window and no data loss.
What changed because of this work?
p99 latency fell from 1.8s to 220ms; infra spend on the feed path dropped 41%.
2022-05 — 2022-11
Owner
What did you do, and why that approach?
Moved 4TB of transactional data across major Postgres versions using logical replication with a scripted cutover and automated rollback.
What changed because of this work?
Cutover took 90 seconds of degraded writes instead of the planned 6-hour outage.
