I build infrastructure that must not fail.
Staff Backend Engineer — Distributed Systems
Most production failures aren't caused where they first appear. Retries, timeouts, and queues often become the visible failure long after an architectural decision was made. Fixing symptoms restores service. Fixing the underlying choice changes which failures are possible.
Selected work
Production
- $2B+ staked · 33k+ validators
Built the backend for MetaMask Staking
Designed the failure boundaries: tenant isolation, idempotent writes, checkpointed recovery, and end-to-end auditability.
Open source
- O(n) → O(1) · ~99% attestations
Fixed the cause of 2-hourly validator restarts in Lighthouse
Removed redundant scrypt runs — every update was re-deriving every key.
- ~26% lower latency
Per-validator MEV-Boost relay routing with hot reload
Lock-free config swap — relays change live without blocking reads.
- 42ms → 311µs · 135×
Removed a 42ms stall on every KurrentDB read
Diagnosed a TCP buffering stall. One socket option, TCP_NODELAY, was all it took.
Essays
On decisions behind systems that have to keep working.
Where reliability matters
The stakes are highest when failure costs money, trust, or reputation.
Remote, London. Staff IC, hands-on, no management.