Senior Site Reliability Engineer by day. I build production control systems for large-scale Kubernetes and cloud environments (AWS 6 regions, 30+ clusters, 600+ nodes). Outside work: AI infrastructure, agentic workflows, and tools that extend what one person can operate alone.

Blog: shaorui0.github.io · GitHub: @shaorui0 · Email: [email protected]

I treat production as a controllable system: make change safer, isolate failure, shorten recovery, and keep efficiency sustainable under SLO guardrails.

• Safer change: progressive rollouts, pause/rollback gates, compatibility preflights (CRDs/webhooks/add-ons).
• Traffic control across clusters/regions: weighted routing for canary, cutover, failover, and staged failback.
• Observability that's actionable during incidents: service-impact signals (latency/error/saturation, SLO burn), incident views that join metrics/logs/changes/deps.
• Cost work with reliability guardrails: Reserved/On-Demand/Spot by workload criticality + interruption tolerance; lifecycle/tiered storage; topology-aware placement.
• On-call recovery: symptom-first triage (refused vs timeout, waiting vs upstream latency, app vs node pressure), smallest-safe mitigation, codified runbooks.

AI Agent Engineering

• Built an autonomous SRE triage agent as a Claude Code skill, used in daily on-call to auto-investigate production alerts via structured debug trees and MCP tool calls (VictoriaMetrics, Grafana, Loki).
• Designed 3-layer code-level safety system — query validation, cluster-tier enforcement, scope tracking — ensuring all agent operations are read-only; unknown environments default to the most restrictive policy.
• Engineered a 132-file knowledge base (6-cluster routing, 7 executable debug trees, 35 incident cases) with JSONL observability tracing for every tool call and safety decision.

Infrastructure & Reliability (AWS 6 regions | K8s 30+ clusters | 600+ nodes)

• Led production Kubernetes upgrades (1.24→1.29) across 30+ clusters; built automation tool cutting per-cluster time from 18–21h to 3–4h with zero incidents.
• Designed distributed auto traffic-switch system across 4 AWS/K8s backends, reducing failover from manual 5–15 min to seconds.

Observability & On-call (40 tenants | ~1.2M active series)

• Built full-stack monitoring (VictoriaMetrics + Grafana + Loki) serving 30 clusters and 40 tenants; migrated from Prometheus Federation, eliminating weekly OOM and reducing data lag from 45s to <5s.
• Designed 3-tier alerting with per-tenant SLI rules, enabling tenant-level fault isolation and SLA tracking.
• Primary on-call for 30+ production clusters; resolved 20+ P1/P2 incidents with avg MTTR ~30 min.

Developed Go microservices and Kubernetes operators for platform tooling; owned production monitoring infrastructure and collaborated with product teams to design end-to-end observability.

• Tencent (Software Development Engineer Intern, 2021/05–2021/09): Go gRPC microservices for live-streaming workloads (Kafka, MySQL, Redis).
• ByteDance (Software Development Engineer Intern, 2021/03–2021/05)
• App Annie (Web Backend Engineer Intern, 2020/11–2021/02)
• Baidu (Data Engineer Intern, 2020/01–2020/10)

• shaorui0/agents — SRE Triage Agent (Anthropic SDK reimplementation with full agent loop, safety gates, eval framework)
• shaorui0/context-infrastructure — Personal AI-native context system, memory architecture, and agentic workflow infrastructure
• shaorui0/chatgpt_flow — Browser extension to extract and organize ChatGPT conversation data

• Languages: Python, Go, Java, Shell
• Infra & data: AWS, Kubernetes, Helm, Docker, MySQL, ClickHouse, Redis, Kafka
• Observability & platform: Prometheus, VictoriaMetrics, InfluxDB, Grafana, Loki, Jenkins, Ansible
• Human languages: Chinese, English

• Master — University of Science & Technology Beijing, Computer Science and Technology (2019/09–2022/06)
• Bachelor — Wenhua College, Electronic Information Engineering (2014/09–2018/06)

Outside of production work, I spend a lot of time at the intersection of AI and infrastructure:

• Building agentic workflows and multi-agent systems — not just using LLMs but wiring them into real operational loops
• AI-native tooling: context systems, memory architectures, orchestration patterns (heavy Claude Code / OpenCode user and contributor)
• Writing about SRE mental models, AI collaboration, and what it actually takes to operate production systems at scale