Networks. Cloud. Possibility.

Jhonny
Macias
Garcia

Network & Cloud Specialist
| Infrastructure Engineer & Home-Lab Architect

I build robust digital infrastructures, from bare-metal setups to cloud services. My home lab is where curiosity becomes reliable, secure systems.

ITS Academy Angelo Rizzoli

Hands-on stack

01 / The person behind the systems

A practical mind. An
infrastructure mindset.

Learning the theory.
Building the real thing.

I’m a Network & Cloud Specialist student at ITS Academy Angelo Rizzoli, following the VET course and working toward graduation in July 2027. Enterprise-grade networking, cloud technologies, and dependable infrastructure are my focus.

I learn by designing, deploying, breaking, and improving systems. My home lab turns abstract concepts into practical experience with virtualization, access control, observability, and recovery.

Adaptable. Detail-oriented. Operations-ready.

My background in technical and administrative operations has taught me to adapt quickly and work methodically. I’m eager to bring that mindset to data center operations and infrastructure teams.

02 / Validated knowledge

A foundation built to keep growing.

AWS Certified Cloud Practitioner badge Achieved

Amazon Web Services

AWS Certified Cloud Practitioner

CLF-C02 · Cloud fundamentals

May 2026 · Valid to 2029 Verify
Microsoft Certified Fundamentals badge Achieved

Microsoft

Microsoft Azure Fundamentals

AZ-900 · Hands-on with Azure VMs and VNets

May 2026 Verify
Microsoft Certified Fundamentals badge Achieved

Microsoft

Microsoft Security, Compliance, and Identity Fundamentals

SC-900 · Security foundations

Jul 2026 Verify

Training badges

AWS Academy Cloud Foundations Trained badge Earned

AWS Training & Certification

AWS Academy Graduate – Cloud Foundations

Training badge · Cloud concepts, core services, security, pricing

Apr 2026 Verify
Microsoft Datacenter Academy Introduction to Datacenters Course Graduate badge Earned

Microsoft Datacenter Academy

Introduction to Datacenters – Course Graduate

Foundational · Datacenter infrastructure and operations

Mar 2026 Verify

Next on the roadmap

Next goal

Linux Foundation

Linux Foundation Certified System Administrator

LFCS · Linux administration

Next goal

Cisco

Cisco Certified Network Associate

CCNA · Routing & switching

03 / Built, not just studied

Small footprint. Enterprise thinking.

What started with an ISP FastGate router has evolved into a fully virtualized, enterprise-grade home lab. Workloads are deliberately distributed: heavy compute on the main host, critical network services on a dedicated node, and backups on isolated hardware.

  • Proxmox VEProxmox Backup Server

    Two-node Proxmox cluster

    Geekom main host and HP backup node, kept quorate by a Raspberry Pi 4 QDevice.

  • OPNsenseCloudflare

    Segmented perimeter

    WAN, LAN, and DMZ with deny-by-default rules, GeoIP filtering, and admin access only via LAN or VPN.

  • AuthentikVaultwarden

    Centralized identity

    Authentik as single source of truth, with forward auth and layered 2FA.

  • PrometheusLokiGrafanaMake

    Observability & AI alerts

    Metrics and logs from hosts, containers, and guests to Grafana on Oracle Cloud, with AI-enriched alerts.

Explore the full architectureShow less

Distributed nodes. Deliberate workloads.

01. Main Proxmox VE node

Proxmox VEOPNsenseDocker

Geekom A6 Mini PC

AMD Ryzen 7 6800H · 16 GB RAM · 1 TB NVMe SSD (OS) + 1 TB external HDD

The main compute host runs all heavy virtual machines, backend containers, and the OPNsense firewall and gateway. Proxmox VE provides the foundation for a fully virtualized environment: the OS lives on the 1 TB NVMe SSD, while a 1 TB external HDD stores Immich’s photo and video library.

Together with the HP notebook it forms a two-node Proxmox cluster, with the Raspberry Pi 4 acting as QDevice to keep quorum.

  • Proxmox VE
  • Two-node cluster
  • Heavy VMs
  • Backend containers
  • OPNsense

02. Network node 24/7

Raspberry PiAdGuard HomeWireGuardRustDesk

Raspberry Pi 4

2 GB RAM · 120 GB SSD

AdGuard Home provides DNS and DHCP, WireGuard secures VPN access, RustDesk handles remote support, and the Pi is the cluster’s QDevice. Prometheus — alongside cAdvisor and pve-exporter — collects the lab’s metrics and Loki stores its logs, both queried by Grafana in the cloud.

  • DNS
  • VPN
  • QDevice
  • Prometheus
  • Loki

03. Isolated backup node

Proxmox Backup Server

HP Notebook

4-core CPU · 8 GB RAM · 1 TB HDD · Headless

The second Proxmox cluster node runs Proxmox Backup Server, keeping secure, isolated incremental backups of the main Geekom node. Scheduled backup jobs capture the workloads, prune jobs enforce the retention policy, and garbage collection reclaims space from expired chunks in the deduplicated datastore. Recovery is a separate responsibility, not an afterthought.

  • Proxmox Backup Server
  • Incremental
  • Retention
  • Garbage collection

04. Development workstation

Apple macOS

MacBook Air M3

Development + local virtualization

The MacBook Air M3 is the development workstation for the lab, supporting hands-on engineering and local virtualization alongside the dedicated infrastructure nodes.

  • Development
  • Virtualization

05. Cloud monitoring VM

Oracle CloudGrafana

Oracle Cloud · Grafana

Grafana only · Private VPN connection

Grafana is the only service hosted on this cloud VM. It queries Prometheus and Loki running on the Raspberry Pi 4 at home through the private WireGuard VPN, and its database is backed up daily with a tested restore.

  • Oracle Cloud
  • Grafana
  • WireGuard VPN

Security isn’t a layer added later.

Perimeter security

OPNsense

A considered perimeter.

OPNsense

The network is split into three segments — WAN, LAN, and a DMZ that hosts the Proxmox guests — with DMZ-to-LAN traffic denied by default.

Hybrid NAT and GeoIP filtering block malicious inbound traffic from high-risk regions before it reaches any service.

  • Firewall
  • DMZ
  • GeoIP filtering

Identity & access

AuthentikVaultwarden

One source of identity.

Authentik + Vaultwarden

Authentik acts as a Single Source of Truth for centralized authentication across the lab’s services.

It also guards Vaultwarden: the web vault sits behind Authentik and its 2FA, then Vaultwarden enforces its own 2FA, which protects the mobile and browser apps too.

  • SSO
  • Forward auth
  • 2FA

Reverse proxy

Nginx Proxy ManagerCloudflare

Private by intention.

Nginx Proxy Manager + Cloudflare

Admin interfaces such as Proxmox, OPNsense, and NPM are reachable only from the local LAN or WireGuard VPN, enforced by strict Access Lists.

SSL certificates use the custom jhonnylab.com domain managed on Cloudflare, and user-facing services are published through Cloudflare behind Authentik.

  • SSL
  • Access Lists
  • LAN/VPN only

Multimedia & storage

Immich

My photos. My infrastructure.

Immich

Self-hosted photo management and synchronization, backed by a dedicated 1 TB external HDD on the main node, provide a practical workload for storage, service deployment, and maintaining control of personal media.

  • Self-hosted
  • Photo sync
  • Storage

Governance & monitoring

PrometheusLokiGrafanaMake

Visibility before guesswork.

Prometheus · cAdvisor · pve-exporter · Loki · Alloy · Grafana on Oracle Cloud

Prometheus on the Raspberry Pi 4 collects three layers of metrics — physical nodes, Docker containers, and Proxmox guests — and Grafana on Oracle Cloud queries it over the WireGuard VPN.

Grafana Alloy ships the system journal of both Proxmox nodes and the Pi to Loki, so logs sit next to metrics. Log-based rules flag failed backups, repeated SSH login failures, and hosts that go silent.

Alerts flow into a Make workflow, where an AI agent adds context and writes an HTML email with suggested actions.

cAdvisor covers every Docker container on the Pi and listens on localhost only; pve-exporter covers every Proxmox VM and LXC through a read-only API token. Dashboards are organized by hosts, containers, guests, and logs, with tailored alert rules for each layer.

  • Host metrics
  • Container metrics
  • Proxmox guests
  • Centralized logs
  • Private VPN
  • AI-assisted alerts

This is an architecture showcase, not an open service directory. Admin interfaces are reachable only from the local LAN or WireGuard VPN, and user-facing services sit behind Cloudflare and Authentik.

04 / Enterprise labs

The Microsoft stack,
end to end.

The home lab runs on open source. At ITS Academy Angelo Rizzoli I designed and operated, as an individual project, a simulated enterprise network on the Microsoft stack — focused on business continuity, high availability, and centralized identity. Every service was built, tested under failure, and recovered.

  • Windows Server

    Nested Hyper-V datacenter

    Multiple hosts on one workstation, with Live Migration and Hyper-V Replica tested for real.

  • Windows Server

    Redundant identity

    Two Domain Controllers with AD replication, Group Policy, and Kerberos delegation.

  • Resilient core services

    Centralized DNS and DHCP in failover, with scopes and reservations.

  • PowerShell

    Headless administration

    Server Core managed remotely via PowerShell Remoting, plus file services and native backup.

05 / Failover clustering

Failover clustering,
on a single host.

A three-node Windows Server 2022 Failover Cluster built as nested Hyper-V hosts on my main Proxmox node. The network is designed in Proxmox SDN, shared storage is served over multipath iSCSI, and every traffic class lives on its own isolated network.

  • Proxmox VE

    Six isolated networks

    One Proxmox SDN zone: management, heartbeat, two iSCSI paths, two Live Migration paths.

  • Multipath shared storage

    iSCSI with MPIO over two dedicated networks, a disk witness, and a Cluster Shared Volume.

  • Windows Server

    Failover, tested

    Validated cluster, Live Migration, and a VM restarted on a surviving node after a host power-off.

  • Zabbix

    Monitoring by design

    Active Zabbix agents push out through SNAT, so no inbound NAT rules were needed.

06 / Hybrid disaster recovery

When the whole site fails,
the cloud takes over.

A nested three-node Proxmox VE + Ceph hyper-converged cluster at home, replicated to a Proxmox node on Azure over a site-to-site VPN. Container disks are mirrored with two-way Ceph RBD mirroring, and failover and failback were tested without data loss.

  • Proxmox VECeph

    Hyper-converged cluster

    Three Proxmox VE nodes with Ceph and HA, proven with a hard node power-off.

  • OPNsenseAzure

    Site-to-site VPN

    Route-based IPsec between OPNsense and the Azure VPN Gateway, behind ISP NAT.

  • TerraformAzure

    DR site as code

    Terraform builds the network, routing, and a Proxmox VM running its own Ceph cluster.

  • Failover and failback

    Snapshot mirroring every 15 minutes; planned moves in both directions with zero data loss.

07 / Agentic operations

Agents run the checks.
I keep the controls.

The DR lab is operated through Claude Code: an orchestrator delegates infrastructure review, health checks, cost tracking, and failover preparation to four specialised agents. The point is the control around them, enforced outside the model, and failover stays in human hands.

  • Orchestrator and four agents

    One session delegates to an IaC reviewer, a health checker, a cost watcher, and a DR operator.

  • Three autonomy tiers

    Reads run alone, reversible changes need my approval, destructive actions are never an agent's.

  • Guardrails outside the model

    Permission rules, a pre-execution hook, and a sudo whitelist on every node.

  • Least privilege

    A dedicated key-only user and narrow scripts instead of general tools.

08 / The next iteration

Always a work in progress. By design.

The next stage is about making the architecture more resilient, repeatable, and secure. I’m scaling the lab toward enterprise-level practices while continuing to learn from every deployment.

01 / Redundancy

Veeam

Resilience by design.

Redundancy · Veeam · UrBackup

Build on the two-node cluster and its QDevice to scale toward enterprise-level redundancy, reduce single points of failure, and make recovery an integral part of the architecture.

Veeam and UrBackup are next, extending backup beyond Proxmox Backup Server and Windows Server Backup.

Next / Reliability, continuity, and backup

02 / Automation

Ansible

Repeatable, not repetitive.

Ansible · Terraform

Deepen automation with Ansible, building on the Terraform behind the DR lab, to bring greater consistency to provisioning, configuration, and the everyday management of services.

Next / Infrastructure automation

03 / Continuous improvement

Security keeps evolving.

Stronger security practices

Continuously expand security practices, refine access boundaries, and strengthen the operational habits that keep infrastructure dependable.

Next / Hardening and governance

09 / Let’s connect

Good infrastructure starts
with a conversation.

Interested in networking, cloud infrastructure, or the next challenge? I’d be glad to connect. Send a message or reach me directly.

Emailjhonny.macias@itsrizzoli.it LinkedInJhonny Macias Garcia GitHubJhonnyJGarcia

Send a message

All fields are required.

04 / Enterprise labs

Nested virtualization. Real failover.

01. Virtualization & business continuity

Windows Server

Downtime as a design constraint.

Hyper-V · Nested virtualization · Live Migration · Hyper-V Replica

Multiple Hyper-V hosts ran as nested VMs on a single physical workstation, turning one machine into a multi-host datacenter where Live Migration and Hyper-V Replica could be exercised for real.

Running VMs moved between hosts without being shut down, and planned and test failovers validated disaster-recovery scenarios end to end.

VM lifecycles were managed through Import and Export, with resources sized deliberately: vCPU allocation, Dynamic Memory, and VHDX virtual disks.

  • Hyper-V
  • Nested virtualization
  • Live Migration
  • Hyper-V Replica
  • Disaster recovery

02. Continued at home

Failover Clustering.

WSFC · iSCSI · Cluster Shared Volumes

Without shared storage, availability came from Live Migration, Replica, and redundant services. I took the next step in my own lab: a Windows Server Failover Cluster on iSCSI shared storage, with a Cluster Shared Volume and a quorum witness.

Quorum and node membership are already familiar ground: my Proxmox two-node cluster relies on a QDevice to stay quorate.

Built / See the cluster lab

Identity & directory

Windows Server

Two DCs. One directory.

AD DS · Kerberos · Group Policy

Two Domain Controllers in redundancy keep authentication available and replicate the Active Directory database between them.

Users, computers, and delegated permissions are managed centrally, Group Policy enforces configuration across the domain, and Kerberos constrained delegation secures Live Migration and Replica between hosts.

  • AD DS
  • Redundant DCs
  • Group Policy

Core network services

Leases that never stop.

DNS · DHCP failover

Centralized DNS and DHCP serve the whole domain, with DHCP in a failover relationship between two servers so leases keep flowing if one goes down.

Scopes handle dynamic clients, while reservations pin addresses for servers and infrastructure.

  • DNS
  • DHCP failover
  • Reservations

Administration & backup

Windows ServerPowerShell

Headless by default.

Server Core · PowerShell Remoting

Servers ran both Server Core and Desktop Experience, with Hyper-V and server roles administered remotely from another domain node.

Management went through PowerShell Remoting and remote consoles rather than local sessions. File services covered SMB shares, Shadow Copies for self-service recovery, Data Deduplication, and Windows Server Backup.

  • Server Core
  • PowerShell
  • Backup

05 / Failover clustering

Network first. Then storage. Then the cluster.

01. Network by design

Proxmox VE

One zone. Six networks.

Proxmox SDN · Simple zone · SNAT · Jumbo frames · RSS

A dedicated SDN Simple zone keeps all cluster traffic on one node and carries six isolated networks: management, cluster heartbeat, two iSCSI paths, and two Live Migration paths. The Domain Controller sits on management and both storage networks; the three Hyper-V nodes see every network.

Only management has a gateway with SNAT, which gives the servers time sync and a way out to monitoring. The other five networks have no DNS and no gateway, do not register in DNS, and run jumbo frames (MTU 9000, verified end to end with unfragmented pings) with RSS enabled.

  • Proxmox SDN
  • Network isolation
  • SNAT
  • Jumbo frames
  • RSS

02. Shared storage

A SAN on the DC.

iSCSI Target · MPIO · CSV

The Domain Controller doubles as the iSCSI Target Server, reachable over both storage networks. MPIO on every node turns them into redundant paths.

It presents a 1 GB disk witness and a 60 GB disk that becomes a Cluster Shared Volume.

  • iSCSI
  • MPIO
  • Disk witness
  • CSV

Cluster

Windows Server

Validated, then proven.

WSFC · Cluster validation · Live Migration

Disks were initialized on all three nodes, the cluster was created on the first and extended to three, and cluster validation passed before anything ran on it.

Network roles are deliberate: management carries cluster and client traffic, heartbeat and Live Migration are cluster-only, iSCSI is excluded. Live Migration is limited to its two dedicated networks. A highly available VM was live-migrated between nodes and failed over when its host was powered off.

  • WSFC
  • Validation
  • Live Migration
  • Failover

Nested virtualization

Proxmox VEWindows Server

Build one, clone two.

Host CPU · Nested Hyper-V · Sysprep

The first Hyper-V node runs with the host CPU type and nested virtualization enabled, with its roles pre-installed.

It was sysprepped and cloned before joining the domain, so each clone got a fresh identity, a new name, and then its domain join.

  • Nested virtualization
  • Sysprep
  • Cloning

Monitoring

Zabbix

Using the design I had.

Zabbix · Active agents · Docker

A Zabbix server in Docker on my workstation receives data from active agents on the servers, which connect outward through SNAT.

Passive agents would have needed inbound NAT. I chose to reuse the existing architecture rather than add exposure.

  • Zabbix
  • Active agents
  • No inbound NAT

06 / Hybrid disaster recovery

Replicate. Fail over. Come back.

01. Replication & recovery

CephProxmox VE

Off-site, and back again.

Ceph RBD mirroring · Snapshot mode · Two-way · Single-node DR cluster

Each disk is primary on one site at a time. Every 15 minutes the primary takes a mirror snapshot, and an rbd-mirror daemon on the other site pulls only the changed blocks through the VPN. A stretched cluster was ruled out: Corosync and synchronous Ceph writes need LAN latency.

A planned failover demotes one side and promotes the other, so nothing is lost. A disaster calls for a forced promote, and the resulting split-brain is resolved with a resync when the home site returns. The DR cluster runs on one VM, so a dedicated CRUSH rule keeps its two replicas on separate disks.

  • Ceph
  • RBD mirroring
  • Failover
  • Failback
  • RPO / RTO

02. Learned the hard way

A corruption, explained.

Kernel log · Root cause · Fix

The first two-way failback corrupted a test container: it started on a disk that was not yet primary while the mirror daemon was still writing to it.

I traced it through the kernel log. Promote, check, and start are now one chained command, and the repeated test passed.

  • Troubleshooting
  • Root cause

Azure foundation

TerraformAzure

Private by default.

Terraform · VNet · Route table · Azure Policy

The DR node has no public IP: it is reached only through the VPN. Failed-over containers live on a Proxmox SDN network routed by an Azure route table, with SNAT for outbound traffic.

Budget alerts and Policy-enforced regions and VM sizes kept the lab inside the trial credit.

  • Terraform
  • UDR
  • Cost guardrails

Agentic operations

Operated by agents.

Claude Code · Sub-agents · Human in the loop

Health checks, cost tracking, and failover preparation for this lab are delegated to specialised agents, with limits enforced outside the model.

See the agentic lab

Documentation

Written up as a guide.

GitHub · Runbooks · Tested results

Decisions, pitfalls, and test results became an open guide to off-site DR for Proxmox and Ceph, alongside the lab's Terraform code and failover runbooks.

Read the guide on GitHub

07 / Agentic operations

Delegate the work. Keep the decisions.

01. Roles & tiers

One orchestrator, four specialists.

Claude Code · Sub-agents · Sonnet & Haiku

I talk only to the orchestrator, whose rules live in the repository: roles, autonomy tiers, and when to escalate. It delegates to an IaC reviewer for Terraform changes, a health checker for VPN, Ceph, and mirroring on both sites, a cost watcher for the trial credit, and a DR operator that prepares failovers. Lighter models read status against fixed thresholds; stronger ones reason across several outputs.

Failover and failback stay in tier 3 by choice: they are rare and have two irreversible points. The DR operator runs the pre-flight checks and writes the exact procedure; I run it.

  • AI agents
  • Autonomy tiers
  • Human in the loop
  • Escalation

02. Enforcement

A prompt is a request, not a control.

Permission rules · Hook · sudoers

Each tier is enforced in three independent layers: permission rules that allow, ask, or deny each command; a pre-execution hook that also reads commands sent over SSH; and a sudo whitelist on every node.

The node's whitelist is the real boundary: it refuses anything off the list, whatever runs on my machine. Agents cannot edit their own guardrails.

  • Least privilege
  • Defense in depth

Narrow interfaces

Specific commands, not general tools.

Read-only scripts · Input validation

Cost data comes from a read-only script instead of a general API tool that could also write. Container details come from a root-owned helper that accepts exactly one numeric ID, after I found that the obvious sudo rule would have leaked a key.

  • Security
  • Azure Cost Management

In practice

Useful from the first run.

Real findings · No invented figures

The health checker found a real Ceph warning on its first run, traced to a pool on the wrong placement rule. Asked about costs before the cost watcher existed, the orchestrator said what it could not verify instead of inventing a figure.

  • Troubleshooting
  • Honest reporting

Next steps

Known limits.

Service principal · Scheduled checks

Azure commands still run under my own login, so the next step is a read-only service principal for agent sessions. Agents also run only on demand: a scheduled health check would watch the lab overnight.

See the lab on GitHub