Nguyen Hong QuangDevOps / AIOps EngineerHa Noi, Vietnam
GitOps for 9 engineering teams, an on-premise GPU/LLM cluster at 99.9% uptime*, and AI agents that cut operations toil by 30%*. Previously Tech Lead AI at KienlongBank.
5,000+
concurrent users the platform is designed for (capacity target), government first
100
ArgoCD applications kept Synced / Healthy for 9 teams
28% → 0.18%
OCR job failures after one root-cause fix
−30%
cloud compute cost at KienlongBank through FinOps
*Uptime and operator-workload figures are operator-reported from internal tracking, not yet instrumented. Every other number on this site comes from benchmarks, HAProxy statistics, Prometheus or Git history, with the source named in each case study.
// git log --first-parent career
From one bank's internal AI platform to a national-scale platform for government departments. The scope grew; the operating model changed.
2026-06 → now
Ha Noi
Own infrastructure for a national-scale AI platform built for 5,000+ concurrent users, rolled out to central and provincial government departments. Direct GitOps delivery for ~9 engineering teams, run the GPU/LLM serving cluster and the OCR platform, and built a digital workforce of AI agents that cut operator workload by about 30%.
2024-12 → 2026-05
Ha Noi
Led the AI team and built KILOBA, the bank's enterprise AI platform: 10+ chatbots and services from research to production, 15+ microservices on AWS and GCP for thousands of staff. Zero-downtime migration to a new AWS Organizational Unit, cloud compute cost down ~30% through FinOps, deployments from days to minutes with GitOps.
2024-02 → 2024-11
Site-to-site VPNs to AWS VPCs, incident troubleshooting on CMC Cloud and AWS, monitoring with EFK and Prometheus; AWS lab environments for hundreds of CCNA, CCNP and AWS SAA trainees.
2018 → 2023
// ls case-studies/
Four production systems, each told the same way: the problem, the architecture, the decisions, and the numbers that came out the other side.
01 · C-OpenAI · 2026
Two-tier routing over an on-premise GPU cluster, a recurring GPU crash root-caused and removed, a new node live in 53 minutes.
02 · C-OpenAI · 2026
One AI agent per platform component, a manager agent, and a hard human gate on every production change.
03 · C-OpenAI · 2026
A silent scheduler failure found and fixed, a two-tier load-balanced GPU cluster built, and a network design corrected the same day.
04 · KienlongBank · 2024–2026
10+ AI services from research to production, a zero-downtime AWS OU migration, GitOps delivery and a FinOps program.
// what I build
GitLab CI → image-tag bump → ArgoCD app-of-apps. Standard router pipelines for dev, staging and prod that nine teams ship through every day.
vLLM behind LiteLLM and HAProxy two-tier routing; benchmarked with guidellm and AIPerf, monitored with Prometheus, Grafana and GPU exporters.
AI agents, one per platform component, coordinated by a manager agent. They diagnose, propose GitOps changes and report daily; I approve every production change.
Vault (OpenBao) secrets with auto-unseal and reload, hardened non-root images, and SOPs for release, change, incident/RCA and backup/DR.
// cat skills.yml
AWS (EC2, EKS, VPC, RDS, S3, IAM, Organizations), GCP (Vertex AI), Kubernetes, Helm, Docker, Docker Swarm
GitLab CI/CD, ArgoCD, AWS CodePipeline, Terraform, Ansible, Vault / OpenBao, Kubernetes RBAC, OIDC SSO
Prometheus, Grafana, Loki, Tempo, OpenTelemetry, Mimir, Alloy, EFK
vLLM, LiteLLM, NVIDIA Triton, GPU clusters, RAG / LightRAG, k6, guidellm, AIPerf, Python, Bash, n8n, Claude Code
Open to senior DevOps, SRE and platform roles where AI workloads run in production.