Spotify

01

A capacity tracker built solo, and used to plan the peak season

Seeing the compute the fleet can actually get used to mean jumping between many monitoring systems and panels in GCP. I built a single pane of glass for it, alone. It shows what compute is available, reserved and running out, per region, live, for anyone in the company who needs it. Everyone and everything is visible in one place, quickly, which was basically impossible before. It was picked up the moment it existed and now drives compute planning for the rest of the year, critically for Black Friday and Cyber Monday.

Built by

Solo

Stack

Java, React, Grafana

02

A service that can take any machine out of the fleet, and never a whole cluster

At this scale the risky part of removing an unhealthy machine is doing it without causing an outage. Every node in the fleet runs a health agent I built, and nodes are never allowed to retire themselves. They report to one central service, where I built the part that decides and acts. It takes a faulty node out of rotation, keeps it for debugging and alerts the team, and it is rate-limited so it can never take down a whole cluster. A service that can reach every machine in a global fleet is only as safe as its limits, so the limits were the work.

Reach

Every node, globally

Stack

Python, Bash, Java, gRPC

03

The agent decides, the scripts act

Upserting a custom tenant cluster touches GitHub Enterprise, Backstage and Jira. I automated it so AI agent skills make the decisions, and every interaction with those systems goes through strict Python scripts the agent is required to use. Whatever gets created or posted is deterministic and consistent, however the agent got there.

04

Upgrades and handovers across the whole platform, from the first days

Alongside the build work I upgraded critical platform services across Spotify's global server fleet, policy enforcement among them, and that work has continued with image verification. I also took part in the global GKE version upgrades and took over several projects in handover, among them a new multi-tenant auth and IAM system and the next generation of service auth. None of it waited on a ramp-up. I was working across the entire platform, globally, within the first days.

Regions

EU, US, Asia and more

Since

September 2026