Maya ChenStaff Software EngineerSeattle, WA
Demo
View PDF

Professional Summary

Builds distributed control planes for fleets that cannot pause for a region outage. Placement, lease management, and API deprecation are the day job; the night job is teaching other teams how not to invent a second source of truth.

At Northline, led the scheduler that cut cross-region failover from 90 seconds to 11 across 14 clusters. Before that, shipped Harborline's mesh and shard rebalancer, and learned packet-level telemetry at Rainier Packet Co.

Skills

Group by:
12 years:
Distributed Systems
10 years:
Go
AWS
9 years:
Observability
Kubernetes
8 years:
System Design
TypeScript
7 years:
Technical Leadership

Work Experience

Northline Control Systems
Seattle, WA, April 2021 to Present
Seattle infrastructure studio that sells a managed control plane to mid-market SaaS. The product is the scheduler; the support burden is everyone else's YAML.

Staff Software Engineer Northline Control Systems

Designed the scheduler that places stateful workloads across 14 clusters and fails a zone in 11 seconds, down from 90. Placement is a lease plus a drain, not a hope that DNS will catch up. Cut dual-write incidents by 73% in the first two quarters, when product teams stopped opening a second writer.
Replaced ad-hoc heartbeats with a quorum lease API that membership actually honors during a partition. Leader flaps dropped 81% and on-call pages for split-brain went from weekly to two in a year. The old ping loop is still in the repo, commented, as a warning.
Versioned the public CRDs and shipped TypeScript/Go clients so product teams stopped scraping etcd for cluster state. Adoption reached 19 internal teams in six months, which is when the deprecation calendar stopped being a slide. The sunset date held; the unofficial watchers did not.

Harborline Software
Seattle, WA, August 2016 to March 2021
B2B SaaS platform that grew from a single-region monolith into a mesh it did not plan for. Maya owned the parts that broke first.

Senior Software Engineer — January 2019 to March 2021Harborline Software

Moved 46 services onto a sidecar mesh with staged mTLS and a sunset date for the plaintext path. p99 inter-service latency fell 28% after retry and timeout defaults were made explicit instead of inherited from whoever copied the last YAML. On-call stopped treating timeouts as a personality trait.
Wrote the Go rebalancer that drained hot shards without pausing writes, which is the part the Friday-afternoon version never quite did. Peak hotspot CPU dropped from 94% to 61% during Black Friday traffic. The runbook is three commands; the design doc is why we did not pause the world.

Software Engineer — August 2016 to January 2019Harborline Software

Built the event pipeline that turned cluster-hours into invoices a finance team would actually send. Late events fell from 4.2% to 0.3% after exactly-once keys landed and the 'we will reconcile later' path was retired. Billing disputes dropped with the late events, which was the real product requirement.
Added drain-aware hooks so scale-in stopped killing in-flight control loops mid-reconcile. Node replacement incidents dropped 62% in two quarters, and the 'why is the leader gone' page stopped pairing with a scale-in event. The hook is boring; the incidents it prevented were not.

Rainier Packet Co.
Redmond, WA, July 2014 to July 2016
Redmond networking shop that sold packet classifiers to regional ISPs. Small enough that every engineer touched the datapath.

Software Engineer Rainier Packet Co.

Ported the rule compiler from a Perl pipeline to Go so a 10 Gbps classifier could reload without a coffee break. Classification latency fell 41% and rule-reload time went from 18 seconds to 2. The NOC noticed the reload first; marketing noticed the latency later.
Shipped the first structured exporter the NOC would trust instead of a packet dump and a guess. MTTD for blackhole routes improved from 22 minutes to 6, which is the difference between a customer ticket and a page we already had. The schema is still the one they page from.

Featured Projects

Northline multi-region placement engine design notes

Open design notes and a reference scheduler that places stateful workloads across regions and drains a failed zone in 11 seconds. Built for operators who are tired of a second etcd becoming the architecture. Includes the lease rules that kept dual-write incidents from returning with the next region.

Distributed Systems
Kubernetes
Go
System Design

Production control-plane SLO and paging playbook

A public write-up of the reconcile-latency and leader-election SLOs Northline uses in production, including the dashboards that made split-brain pages rare. Written so a new on-call can tell a burn from a noisy saturation without a war room.

Observability
Technical Leadership

Education

University of Washington

B.S. Computer Science — June 2014

Certifications

Certified Kubernetes Administrator

Cloud Native Computing Foundation — April 2020

Credential ID: CKA-MAYA-8841

AWS Certified Solutions Architect – Professional

Amazon Web Services — September 2022

Credential ID: SAP-MAYA-2209