Skip to content

[ Platform engineer · AI infrastructure ]

The infrastructure under the agents, not the prompts on top

Almost everyone building for AI agents arrives from the application side. I arrive from the metal — Kubernetes on hardware I own, mesh, split DNS, GitOps that reconciles in under thirty seconds. That platform is where my own agents now run, under a harness with hard limits. The homelab below is the proof. The agent platform is what I am building on it.

Explore the platform → What I build Tell me what you are building

[ What I build ]

KubeLab

8 machines, 3 K3s clusters

I wanted to understand how platforms actually work — not just click buttons on managed K8s. So I built one from scratch. Eight machines, three single-node K3s clusters, VPN mesh, SSO, full observability. Everything IaC, nothing manual.

Kubernetes Go Python Terraform Ansible View project →

Hive

67–82% fewer tokens

AI assistants waste tokens loading context they don't need. Hive queries your Obsidian vault on demand via MCP — only what's relevant, when it's relevant. 67-82% token reduction.

Python MCP Obsidian AI View project →

Pollex

3s on-device inference

Cloud LLM APIs see everything you type. Pollex polishes your English on a $99 Jetson Nano — grammar, coherence, wording. 3-second inference, zero cloud dependencies.

Go MCP LLM Jetson-Nano View project →

ts-bridge

Zero-install, no admin

Corporate firewalls block VPN clients that need admin rights. ts-bridge tunnels RDP/SSH through Tailscale in userspace — no install, no admin, no traces. One binary.

Go Tailscale Networking View project →

pdf-modifier-mcp

Format-preserving edits

Find-and-replace in PDFs without destroying the formatting. CLI for batch jobs, MCP server for AI agents.

Python MCP PDF View project →

yt-metrics-cli

Any public channel

YouTube Studio analytics are shallow and locked to your own channel. This CLI pulls metrics for any public channel and exports everything.

Python CLI YouTube View project →

kasa-provisioner

20 plugs, bulk-provisioned

Configuring 20 smart plugs one by one through a phone app is painful. This scans the network, updates firmware, and sets schedules in bulk.

Python IoT Networking View project →

dotfiles

1 script, fresh machine

Dev environment that rebuilds itself. Zsh, Neovim, tmux, Git — one script on a fresh machine.

Shell Neovim DevEx View project →

[ Platform telemetry ]

Argo CD v3.4.1 · 2 applications synced

KubeLab Hybrid Cloud & Edge Platform

8 Nodes Active 35 Services

K3s Operational · GitOps Synced (ArgoCD)

K3s v1.34.4+k3s1 · Hybrid multi-node fleet

Jetson Nano · Qwen 2.5 1.5B, CPU inference

Tailscale mesh (Headscale) · Hardware inference

Hive MCP Active (Context RAG)

FastMCP context reduction & memory graph

99.9% Uptime (90d)

<30s reconciliation · Zero manual ops

Explore the platform → →

[ Latest notes ]

2026-08-27 The Homelab IDP I Planned, and the One I Actually Run 2025-05-26 Network Topology for a Hybrid Homelab-Cloud Setup 2025-05-19 Debugging DNS: My Systematic Approach
All notes →

[ My path ]

  1. 2019 — Present

    Applications & Platform Engineer · Teledyne E2V · Spain / USA

    Built an Internal Developer Platform on Kubernetes + AWS — infra cost −80% ($1.5K→$300/mo), customer onboarding 60→14 days.

  2. 2018 — 2019

    Technical Specialist · Teledyne E2V · Shenzhen, China

    NPI and technical presales for high-speed imaging across APAC; primary engineering liaison between EMEA and Chinese clients.

  3. 2017 — 2018

    R&D Engineer · Teledyne E2V · Grenoble, France

    C and VHDL firmware for vision cameras at sub-10 ms latency; .NET GUIs for automated validation systems.

  4. 2015 — 2017

    Test Engineer · Teledyne E2V · Seville, Spain

    Automated CMOS sensor testing in Python — eliminated 15+ hours of manual work per week and raised throughput.

[ Open source ]

21 public repos
16 stars · GitHub
10 stars · Hive
Shipped in
  • Python
  • Go
  • Astro
  • MDX

[ Community & background ]

Cloud Native Sevilla CNCF
NaN AI builders
FAR Active rower