Enterprise Data & AI
at Home

A private cloud solution based on k8s running a production-grade lakehouse, autonomous agents, and full observability. It is assembled from open source and operated from Git in a home rack.

Explore the 14 projects See how it runs GitHub

Standing on open source

The platform is assembled from battle-tested open source rather than one vendor, from the kernel to the dashboards. These are the foundations the rest of the site describes.

Kubernetes K3s Argo CD Helm Ansible Prometheus Grafana Loki PostgreSQL Valkey Docker Tailscale Ollama Garage Longhorn Superset n8n Polaris

Featured work

Six of fourteen projects, each with a full case study behind it.

Power on the desk

Seven nodes you can touch: two architectures, three storage tiers, and a GPU node in one home rack. Every figure here traces back to the live cluster.

7
Cluster nodes
Architecture
25+
Services deployed
Services
52 cores 140 GB
Compute
Hardware
3
Storage tiers
Architecture
20
Monitoring targets
Monitoring
15
Repositories
GitHub

Read the cloud-at-home case study →

Enterprise practices, not a lab

Every discipline an enterprise expects, running on hardware you can touch.

Infrastructure as code

Ansible provisions the OS and K3s; Helmfile ApplicationSets describe the services. Nothing is set up by hand. Provisioning

GitOps delivery

ArgoCD continuously reconciles one Application per namespace, with server-side apply and auto-sync on every commit. Automation

CI/CD & governance

GitHub Actions gates every change and Renovate keeps dependencies current; all work lands through reviewed pull requests. Cost management favours local inference and free-tier models over per-token spend. CI/CD

Identity, secrets & access

Dex turns GitHub OAuth into OIDC for SSO, OAuth2-Proxy protects services, Tailscale gives zero-trust access, and secrets management uses Sealed Secrets mirrored by Reflector. Security

Observability & reliability

Prometheus, Loki, Grafana, AlertManager, and Robusta watch the estate; reliability and backups come from Longhorn replication, Velero, and Kopia. Monitoring

Data & AI platform

The data platform runs Garage, Iceberg, Polaris, and Trino, with streaming on Redpanda and orchestration through Airflow, dbt, and dlt; the AI platform runs local Ollama inference and Sympozium agents. Data Stack · AI

Experience is still the job

AI now writes much of the how. The work that remains is judgment: what to build, what to cut, and what to trust. This platform is where that judgment was practised.

Open-source vendor risk

“MinIO changed its license and removed the open-source images with very short notice. By the time the removal date was close, it became a forced, rushed migration.”
Choose storage a neutral community governs, before you are forced to move. Hard Lessons

Local AI agents

“AI must enrich detection, not replace it.”
Prometheus and AlertManager still own deterministic detection; the model explains and prioritises. SRE agents

Scope and purpose

“Without a deliberate plan for what the homelab was for, it grew into an unmanageable web of services.”
Every added service is another point of failure. Define the purpose first, then cut what does not serve it. Hard Lessons

What's next

Where the platform goes from here.

  • Broader MCP tool coverage. Wider fact-gathering with every mutating operation behind explicit policy and approval.
  • More use cases. New data sources, streaming pipelines, and agents, each added as a card and a case study.
  • Impact metrics. Longer-running outcome measures with a defined source.

Authors

Who built this, and where to find the work.

Alvaro Santos Andres

Alvaro Santos Andres

Data & AI engineer building production-grade platforms on open-source tech. DataHub.local is a real, running homelab. It spans Kubernetes on ARM64, LLM pipelines, and GitOps, all built and improved in the open.

Personal site LinkedIn GitHub
Alvaro Santos Andres