Fleet Automation at Bloomberg

June 2025 – Present

After 3 years of building ESG data pipelines at Bloomberg, I transferred internally to the Fleet Automation Services team: infrastructure where the problem isn't only making a system work, but making it safe to roll out. I help build a Go-based orchestration platform spanning 7 services and more than 100,000 machines, turning operational work into campaigns that are observable, predictable, and safe to run at scale.

Replacing manual router upgrades

I led an expansion of the fleet automation platform to 6,000 Juniper routers. Previously, quarterly upgrades required engineers to connect over SSH, sequence changes by hand, and remain available to diagnose problems — work that could add up to 3,000 engineer-days in a single cycle. With the networking teams, we translated that operational reality into scheduling rules that account for regional maintenance windows, dependencies, concurrency limits, and staged rollouts. The result is a process designed to reduce manual coordination and operator fatigue while keeping safety controls in the rollout itself.

Keeping fleet state current

Fleet automation is a distributed-systems problem: eligibility, scheduling, ticketing, and reconciliation all need an accurate view of where a machine is in its lifecycle. I co-designed the Go platform's evolution from batch status recomputation toward Kafka-driven events, reducing stale state and the race conditions that follow from it. Not every workflow needs to be real-time, so the design keeps scheduled processing where it fits while using events where prompt feedback matters to the people responsible for the machines.

Rolling out security safely

I designed and implemented a security campaign driven by the Digital Operational Resilience Act (DORA) spanning 60,000 machines and 6 infrastructure teams. The work extended fleet orchestration with multi-stage rollouts, telemetry-based prechecks, and campaign-specific scheduling constraints. Rather than treating a rollout as one large switch, the campaign advances through deliberate gates so teams can verify that a machine is ready before it moves forward.

Owning the campaign lifecycle

I've led 2 fleet automation campaigns from infrastructure requirements through technical proposals, architecture, implementation, and production support. That scope makes the work as much about communication and operational clarity as code: clear metrics and logging make production behavior understandable, while close partnership with the teams running the infrastructure makes the automation useful in practice.