whoami --verbose

Weigang Geng

Staff/Senior SRE · AI Infrastructure · Kubernetes, LLM Inference, Cloud

I build and run the infrastructure behind large-scale AI: GPU fleets, Kubernetes and Slurm clusters, and the observability that keeps them honest. 10+ years scaling systems that support $300M+ revenue and 20,000+ GPU clusters. Before that, I was a particle physicist hunting the Higgs boson — I still debug like one.

Impact

3h → <3min
Cut time-to-detect 99% by re-architecting monitoring across lab infrastructure.
Meta
40K → 20 pages/wk
Cut on-call page volume 99.5% with actionable, symptom-based alerting.
Lambda
3,000+ GPUs
Built and operated Kubernetes/Slurm clusters for AI/ML; led first zero-downtime k8s control-plane upgrades.
Meta
80% → 99.9%
Monitoring uptime via Prometheus/Grafana overhaul; 1,500+ dashboards, 300+ services.
Walmart
10,000+ machines
Provisioned physical and virtual machines; 10x faster builds with Puppet/Salt/Ansible.
Machine Zone
800+ nodes
Own reliability for an 8-cluster GPU fleet across five infrastructure providers.
Aranya

Stack

Infrastructure

KubernetesSlurmCUDA OpenStackAWSHPC Kyverno

Observability

PrometheusGrafanaAlertmanager ELKSLO / SLIRoot Cause Analysis

Automation

PythonGoAnsible PuppetSaltStackCI/CD Bash

Certifications: Microsoft Certified — Azure AI Fundamentals  ·  Languages: English (professional), Chinese (native), French (working)

Physics past life

Before infrastructure, I spent a decade in experimental particle physics. I earned my PhD at Michigan State, searched for the Higgs boson on the D0 experiment at Fermilab, and made the world's first measurement of CP violation in single top quarks — presented at the American Physical Society in 2011. I also spent a year at CPPM in Marseille on a French government Eiffel Scholarship, and later did PET brain-imaging research at Oakland University.

The physics never left: root-cause analysis, first-principles reasoning, and a low tolerance for things that "just work by accident." Publications listed on Google Scholar.

PhD, Michigan State Higgs search @ D0 / Fermilab CP violation, single top Eiffel Scholarship, CPPM Marseille PET imaging

Projects

cheatsheets
Personal reference for Linux, DevOps, networking, and ML.
HTML
dsh-kubectl-guard
Policy plugin for kubectl: deny destructive verbs outside local clusters.
JavaScript
agentlens
Observability for AI agents: trace viewer + investigator agent.
TypeScript
skills
Agent skills, installable via npx.
Python
grafana-dashboards
Dashboard generation scripts for Grafana.
Python
grafana-site-weather
Weather dashboard via Open-Meteo & Infinity datasource.
misc