Diagnoses at the component, not the ticket
Thermal bench-testing, firmware and BIOS flashing, soldering and harness wiring. When a node fails I find the part, not the symptom.
Monte Carasso, Ticino. Six years at the bench, five years on Linux, four servers in production. Looking for datacenter operations, hardware integration or Linux systems support, on-site or hybrid.
I build infrastructure from the solder joint up. Component-level diagnostics, ZFS storage, zero-trust networks, and the occasional driver written from a packet dump.
Thermal bench-testing, firmware and BIOS flashing, soldering and harness wiring. When a node fails I find the part, not the symptom.
Ansible playbooks and Docker Compose rebuild the whole environment from a blank disk in under thirty minutes. Nothing important lives only in one place.
Cloudflare Tunnels, WireGuard and Tailscale meshes, SSO with 2FA, host firewalls and intrusion detection on every node. Security is part of the build, not a later phase.
Reverse-engineered a vendor USB HID protocol from packet captures and shipped a working Linux driver upstream. Undocumented hardware is a puzzle, not a blocker.
Upstream contributions to projects that sit between hardware and the kernel. Hover a row and the sculpture becomes the work.
Sniffed the vendor's raw HID reports in Wireshark, decoded the byte layout for DPI profiles, polling rates and on-board memory, then wrote a complete C driver that was merged upstream.
Traced execution crashes in the Steam Linux Runtime to unwritten manifest files, documented a depot recovery workaround and contributed code to automate runtime integrity checks.
Replaced brittle static tables with bounded manifest parsing so NVIDIA DLSS, AMD FSR and Intel XeSS libraries survive DLL upgrades; also fixed UI state desynchronisation.
Four production hosts rebuilt from bare metal by Ansible in under thirty minutes. Twelve-plus daily users, no exposed WAN ports: Cloudflare Tunnels, WireGuard, SSO with 2FA and Fail2Ban on every node.
A 19-inch cabinet for a private client: patch panels, storage chassis and nine drive bays, every cable run dressed and labelled. TrueNAS on top with ZFS RAID 10, per-user ACLs and offsite snapshots for a sub-30-minute recovery target.
Cross-platform provisioning utility for environment configuration and plugin management, adopted by other users.
Coordinated a twelve-person Scrum team building a full-stack city simulation. Owned the Spring Boot backend and its integration with the web frontend.
Also on the bench: Rust, C, C++, Python and SQL for diagnostics, and a self-directed track on local LLM deployment and quantisation on bare-metal hardware with llama.cpp and Claude Code.