HAIYANG LUO

Haiyang Luo / haiyang.dev

I build cloud & compute infrastructure.

Infrastructure and distributed-systems engineer. This page is a descent: from the nebula where I started to the ocean floor.

↓ begin descent

LOW ORBIT

About

Founding engineer at Vybers. I took a venture-backed startup from an empty GCP organization to public production launch, as sole owner of the Terraform-managed cloud infrastructure, the backend services, and the GPU and geospatial platforms.

Before that, I built cloud-native microservices in Go and .NET at Epic Systems and ran production on-call for a platform serving 200+ health systems.

CLOUD LAYER

Experience

Vybers · Founding Engineer

Jan 2026 – Present

  • Built a GPU compute control plane covering scheduling, placement, resource management, and high-availability recovery.
  • Led the platform's three major migrations: Render → GCP/Kubernetes, Python → Node.js, and NestJS built-in TCP → gRPC.
  • Sole owner of the Terraform/GKE foundation, cloud networking, CI/CD, and Prometheus/Grafana observability.

Epic Systems · Software Developer, Nebula

Jan 2023 – Dec 2025

  • Shipped a .NET/Go gRPC service framework adopted by 15+ teams; cut service onboarding time by ~90%.
  • Led a zero-downtime NATS JetStream migration across 200+ production systems.
  • Completed 300+ hours of production on-call for a platform serving 200+ health systems at 99.9%+ uptime.

SEA LEVEL

0 m

MIDNIGHT ZONE

Selected work

GPU compute control plane

Python control plane for multimodal AI inference on GPU capacity: job queueing, placement, lease-based failure recovery, idle spin-down, and cordon-and-drain rollouts. Cut cold starts by ~75%.

Geospatial data platform

PostgreSQL/PostGIS platform serving canonical places, photos, and boundary geometry worldwide; DuckDB-based ingestion at planet scale, with hourly self-healing reconciliation.

Platform migrations

Render to GCP/Kubernetes, Python to Node.js, and NestJS built-in TCP to gRPC with shared Protobuf contracts and a CI guard against regressions, delivered while the product kept shipping.

Cloud infrastructure, 0→1

Custom VPCs, private GKE, service mesh, three-tier Cilium egress policy, keyless CI/CD with Workload Identity Federation, and Prometheus/Grafana observability, from an empty organization to public launch.

THE FLOOR

Let's talk.

You've reached the bottom. If any of this is useful to your team, I'd be glad to hear from you.