Skip to main content
Trevor Grafton

PROJECT / Raspberry Pi lab

Homelab

A Raspberry Pi 5 lab with a local-AI system monitor and checker, plus DNS filtering and an outfit check from my post.

Built
  • Raspberry Pi 5
  • Pi-hole
  • Flask
  • psutil
  • Ollama
Type
Systems
Source
Public on GitHub
Role
Home lab projects: a system monitor, a system checker and a local-AI outfit check (from my post and project folders).
Built with
Raspberry Pi 5, Pi-hole, Flask

01 / Overview

Overview

  1. Collect

    The dashboard reads CPU, RAM, disk and temperature with psutil.

  2. Serve

    A Flask app shows the stats live.

  3. Ask for analysis

    Clicking refresh sends the stats to a local Ollama model.

  4. Read the result

    The model returns a status, an analysis and advice.

  5. Optional alert

    A spoken alert can be played with espeak.

02 / Context and problem

Context and problem

A Raspberry Pi 5 lab. Pi-hole filters DNS for the network and a storage array is planned (both resume-reported). Setup details come from my posts and the recording. The AI system monitor is a public repo, and a related checker adds an agent loop. An outfit-check project uses a Tapo camera and a local Ollama vision model (resume-reported; I do not describe its internals).

03 / My role

My role

Home lab projects: a system monitor, a system checker and a local-AI outfit check (from my post and project folders).

04 / How it works

How it works

  • Pi-hole filters DNS for the network (resume-reported).
  • A system dashboard shows RAM, disk and temperature, and can ask a local model for analysis.
  • An outfit-check project uses a Tapo camera and a local vision model (resume-reported).

05 / Key decisions

Key decisions

AI analysis on manual refresh only

No background model calls, so the dashboard does not spam the Pi.

Small models on the Pi

The monitor uses llama3.2:1b and the checker uses Qwen2.5 Instruct 1.5b, both small enough for the hardware.

06 / Results and numbers

Results and numbers

A small lab at home with two small local-AI tools on it; the dashboard tool is public.

  • The monitor and checker run local models via Ollama; Pi-hole and the outfit check are from my post only.

07 / Limitations

Limitations

  • The two tools use different models, and I have no measured latency.
  • The Pi AI HAT+ did not fit my case (resume-reported).
  • The redundant storage array and backup target are plans only.
  • No automated tests in these tools.

08 / What's next

What's next

  • Build a redundant storage array and a backup target.

09 / Media gallery

Media gallery

Laptop screen recording of the Pi-hole admin page's domain list
Laptop screen recording of the camera used by the outfit check (resume-reported project)
Laptop screen recording of the Pi system dashboard with RAM, disk and temperature
Laptop screen recording of the dashboard's AI Analysis output from the local model

10 / Next project

Next project

Graphify Habit →