I keep AI products running in production.

DevOps & AI Infrastructure engineer in Ho Chi Minh City. Self-taught in tech since 2016, DevOps since 2023.

Do Cao Hieu

Do Cao Hieu, smiling, in a black polo shirt
production servers
5
containers running
123
in tech since
2016
DevOps since
2023

AI-native infrastructure is a different discipline.

Cost

LLM inference costs scale with tokens, not requests.

Data

Vector databases need tuning the way cache layers did a decade ago.

Orchestration

Agent orchestration is replacing cron jobs.

Observability

Now includes prompt regressions, evaluation drift, latency budgets and model behaviour.

I’ve been building and operating AI-native systems since 2025. The goal isn’t deploying a demo — it’s keeping production AI reliable, observable and cost-efficient.

From IT helpdesk to production AI.

The stack I run on.

Cloud & Edge
AWS, Google Cloud, Cloudflare
Containers
Docker, Kubernetes
Infrastructure as Code
Terraform, Ansible
CI/CD
GitLab CI, GitHub Actions, Jenkins
Networking
Traefik, Nginx, Tailscale, systemd
Databases
PostgreSQL, Redis, pgvector, SQLite
Observability
Prometheus, Grafana, Loki, Tempo
Languages
Go, Python, TypeScript, Bash
AI Infrastructure
MCP, Claude Agent SDK, Gemini API

Five servers I run myself.

123 containers across Singapore and Vietnam, one private Tailscale mesh. Measured 26/09/2026.

Cloudflare DNS each hostname → its own node’s Traefik (TLS)
  • server-1Singapore
    traefik · tls
    • devops-lab-platform (sysbox)
    • private apps
    • media-console (public entry)
  • server-2Singapore
    traefik · tls
    • jenkins
    • harbor registry
    • terrakube
    • semaphore
    • postgresql
    • media-console + workers
  • server-3Vietnam
    traefik · tls
    • prometheus
    • grafana
    • loki
    • tempo
    • alertmanager → telegram
  • server-4Singapore · ARM
    traefik · tls
    • gitlab ultimate (self-managed)
    • infrastructure console
    • vault
    • nextcloud
    • beszel hub
  • server-5Singapore
    traefik · tls
    • s3 gateway
    • small web apps

tailscale · one private mesh links all five nodes; each runs independently

status.docaohieu.com Cloudflare Workers, outside the fleet. Watches 33 apps, so it still reports if the fleet is down.
s3 · hot tier
rustfsrustfsrustfs
3 nodes, erasure coding
cold tier
sharepointgoogle drive
long-term copies

Kopia and GitLab backups land here, along with every app’s files.

Things I’ve shipped.

Private tool

Infrastructure Console

My own self-service cloud panel for the fleet: S3 buckets, quotas and per-app keys; exposing a service (DNS + Traefik route, tailnet-only by default); Postgres databases and roles; Harbor projects; one-click Terrakube / Semaphore runs; live fleet status. Passkey-only sign-in, and every change needs a fresh passkey step-up.

Go · WebAuthn · etcdNo public link

Live

Self-hosted S3

S3-compatible object storage: a 3-node RustFS hot tier (erasure coding) over SharePoint and Google Drive cold storage; holds the fleet’s backups and every app’s files.

RustFS · etcd · GoNo public link

Private tool

Media Console

A shared media workbench for my household: download, FFmpeg video and image tools, transcripts, ElevenLabs voice, AI generation and CapCut project export. A Postgres job queue feeds workers on two servers, results land in S3 behind short-lived links, and each person has a spending budget per AI provider.

Go · PostgreSQL · FFmpeg · S3No public link

Private tool

DevOps Lab Platform

My own KodeKloud-style lab platform: browser terminal into a fresh sysbox container per lab, auto-graded tasks.

sysbox · Docker · xterm.js · Next.jsNo public link

Open source

DevOps Automation Toolkits

27 reusable IaC and automation repos: Ansible, Terraform / OpenTofu / Pulumi, ArgoCD / Tekton, CI/CD and monitoring stacks.

Ansible · Terraform · ArgoCD · Tektongithub.com/docaohieu2808 →

Also built: Floatwave — an open-source floating YouTube Music player for Windows (github.com/docaohieu2808/floatwave) · Clinic ERP — scheduling, patient records and billing for a private physiotherapy clinic.

Where I’ve operated.

  1. DevOps Engineer & AI InfrastructureSelf-employed2023 — Present

    • Run 5 production servers with 123 Docker containers over a Tailscale mesh; each node has its own Traefik edge.
    • LGTM observability on a separate node in Vietnam with Telegram alerting; an external status page on Cloudflare Workers watches 33 apps.
    • CI/CD and platform: self-managed GitLab Ultimate, Jenkins, Harbor, Terrakube and Semaphore. Built my own self-service consoles for infrastructure, S3 storage and media, and the DevOps Lab Platform.
  2. AI Data ContributorOutlier2024 — Present

    • Top 5% highest-performing contributor globally.
    • Train and evaluate frontier models (Claude, GPT, Gemini) through expert prompt engineering and code review.
  3. IT & Technical SupportFurniture manufacturer2019 — 2023

    • Ran IT for 30+ staff: workstations, LAN / Wi-Fi, file server / NAS and backups.
    • Maintained CNC machine systems and the production-floor data pipeline.

War stories, written down.

Cover of AI Agents in Production, second edition

Second edition · March 2026 · 13 chapters

AI Agents in Production — War Stories from a DevOps Engineer

A practical guide to building, deploying and operating AI agents in production: model selection, context engineering, cost, observability and security.

Get it on Gumroad

Five free books, in Vietnamese.

Basic to advanced, left to right.

  1. Cover of DevOps Bằng Hình

    DevOps Bằng Hình

    Visual — learn by diagrams

    Read PDF
  2. Cover of Docker Compose Toàn Tập

    Docker Compose Toàn Tập

    12 chapters

    Read PDF
  3. Cover of DevOps Level MAX

    DevOps Level MAX

    25 chapters

    Read PDF
  4. Cover of DevOps Tricks

    DevOps Tricks

    24 chapters

    Read PDF
  5. Cover of DevOps Field Manual

    DevOps Field Manual

    848 pages

    Read PDF

Let’s build reliable systems.

Open to DevOps, Platform and SRE roles — remote or in Ho Chi Minh City.