Giuseppe Castelluccio

In short

Experience

1. IT / systems support

   2026, two weeks

   On-call collaboration, Sicily

   • Maintaining other people’s IT systems and infrastructure, on call.
   • Administration, configuration, fault finding.

2. IT / programming apprentice

   2026, one-month internship

   Computer Mania SRL, Capo d'Orlando (ME)

   • Installed, configured and maintained computer labs for schools.
   • Structured cabling and LAN configuration in an educational environment.
   • Hardware diagnostics: component compatibility, PSU testing, SMART analysis.
   • RMA handling and end-user support; Windows workstation and classroom peripheral setup.

What I run and deliver

1. Bilingual chatbot

   Python · Docker · LLM OpenAI-compatible

   Answers in Italian and English from a business's own information, and hands the enquiry to a spreadsheet, a CRM or an inbox. One version runs offline, the other on an API key. 129 tests. Runs in Docker. I have never installed it for anyone: I wrote it because it is one of the most asked-for things.

2. Five demo websites

   HTML/CSS/JS · nessun framework · generati da uno script

   Five invented businesses, one per trade, each with its own pages in Italian and English and one tool that does a real job. None of them is a client. I built them to have something to show.

3. Three Linux machines I keep running

   Debian 13 · Btrfs · Docker · VMware

   A backup and services box, a virtualization workstation, a laptop. Containerised services, Prometheus and Grafana, self-hosted Git, snapshots replicated off the machine. The configs came from AI. I understand the commands I run, and when it breaks at midnight there is nobody else to call.

4. Offsite backups I have actually restored

   restic · Backrest · Btrfs · SSH

   Deduplicated encrypted backups scheduled and monitored through Backrest, on top of Btrfs snapshots and replicated off the machine over SSH.

5. Local LLMs on consumer GPUs

   Ollama · llama.cpp · CUDA

   Quantized models on my own hardware, including inference split across an RTX 4060 Ti and a GTX 1650 over RPC. A 3.3 GB model answers site-style questions correctly in a few seconds, 3.9 s measured on my own machine, at zero cost per message.

Skills

Systems & OS

Linux (Debian, NixOS, Arch, Void), Windows client; Btrfs snapshots; restic/Backrest backups; bootloaders, partitioning.

Networking & security

LAN, structured cabling, TCP/IP, routing; SSH keys, hardening, forced-command; nftables, AppArmor.

Containers & virtualization

Docker, Podman, self-hosted registry; VMware, QEMU/KVM with libvirt.

Monitoring & services

Prometheus, Grafana, Forgejo.

Code & automation

Reading, configuring and debugging Python and Bash; Git/GitHub; running and interpreting automated test suites; Neovim, VS Code.

AI-assisted delivery

Specifying what software must do, driving AI coding tools to produce it, then testing, deploying and supporting the result on client hardware; local LLMs and OpenAI-compatible APIs.

Hardware

Workstation diagnostics and assembly, PSU testing, SMART analysis, BIOS.