Skip to content

The Lab

What we build for ourselves.

The experiments, tools, and infrastructure we build for ourselves, where we take the risks so your project does not have to. Ideas arrive here as questions on our own hardware and leave as reliable technique. This is where the on-premise expertise comes from.

The homelab

Where on-premise AI actually runs.

This is the hardware behind the pitch. Four machines on a private network, running local AI models on our own silicon. When we say your data never leaves the building, this is the building.

  • cawecapionline

    Raspberry Pi 5

    Sensor node

    ~200MB~2W
    • Environmental sensors
    • GPIO / I2C / Serial
  • serveronline

    Mac Mini M4 · 16GB

    Services & data

    12-14GB~5-8W idle
    • PostgreSQL · Redis · RabbitMQ
    • Django APIs + Nuxt dashboards
    • Caddy reverse proxy
  • brainonline

    Mac Mini M4 Pro · 48GB

    On-premise AI

    32-40GB~8-12W idle
    • Ollama · Gemma · Llama 3.1 · TranslateGemma
    • Rust services
    • AI workers
  • renderstandby

    Desktop · RTX 5090

    GPU render, on demand

    Wakes on demand~0W asleep
    • FFmpeg + NVENC
    • GPU render worker
Private Tailscale mesh No public IPs, no port forwarding Local models, no third-party API ~2-14W idle per machine

The toolbox

What we build with.

AI / ML

  • Local LLMs
  • Fine-tuning
  • RAG & search
  • On-prem inference

Engineering

  • TypeScript
  • Node & edge
  • APIs
  • Databases
  • Testing

Infrastructure

  • Cloudflare
  • Serverless
  • On-premise
  • CI/CD
  • Observability

Design

  • Product (UX/UI)
  • Brand systems
  • Design systems
  • Prototyping

Navegavam sem o mapa que faziam

They sailed without the map they were making

Sophia de Mello Breyner Andresen · Navegações

Internal projects

Independent projectIndependent

A Self-Hosted, Distributed AI Pipeline

A demanding AI and media workload spread across four mismatched machines I already owned, with nothing leaving the network and no meter running.

Local LLMs, translation, and GPU rendering across four machines, surviving node failures mid-job, at zero API cost

  • On-prem AI
  • Distributed
  • Infrastructure
The ClimaSense dashboard in dark mode, showing live temperature, humidity, pressure, and air-quality readings
Independent projectIndependent

Real-Time Environmental Monitoring

A Raspberry Pi sensor platform measuring the actual air in the room, temperature, humidity, pressure, and particulates, with everything from sensor to dashboard on one device.

Live readings from real sensors on a single low-power board, with no account, no cloud, and no subscription

  • IoT
  • Realtime
  • Full-stack

Want this kind of curiosity pointed at your problem?

Start a project See customer work