• Experimented and built Ray clusters using KubeRay, accelerating data-preprocessing speed by 30x.
• Redesigned multiple part of the machine-learning service lifecycle workflow — including cloning and removing features.
• Implemented an automated recovery mechanism that detects unrecoverable service states and triggers immediate restarts.
• Wrote multiple optimization plans and proposals that enhanced service manageability, efficiency, and stability.

• Managed multiple kubernetes clusters with in-depth expertise in cloud platforms.
• Maintained in-house kubernetes operator and open-source kubernetes operator for effortless deployment and management of workloads.
• Refactored the user behavior event pipeline to handle 13k RPS and 120k msg/s traffic efficiently, reducing costs by integrating multiple messaging systems and implemented high-performance batching logic.
• Researched and introduced solutions for observability, resiliency, and performance improvements, including tools like Linkerd, Gateway API, and ArgoCD.
• Administered and optimized the observability stack, including Grafana, VictoriaMetrics, and Tempo.
• Acted as the first responder for incident management, collaborating with cross-functional teams to resolve critical issues.

• Mastered the company's systems within one month, successfully completing the annual PCI-DSS external audit.

gta

GTA is a command-line tool for managing temporary IAM roles across cloud providers

• Automatic role revocation on program exit or interrupt
• Configurable time-to-live (TTL) for permissions
• Unique identifiers for each temporary binding for easy cleanup

go-dataloader

A clean, safe, user-friendly implementation of GraphQL's Dataloader, written with generics in go.

• Written in generics with strong type
• Use shared memory via a communicating pattern
• Provide almost the same interface as the original Node.js Dataloader
• Implement a promise like thunk design, simply call `val, err := loader.Load(ctx, id).Get(ctx)`

go-batcher

This is a batcher inspired by GraphQL's Dataloader and yckao's dataloader implementation.