📍 Newport, South Wales, UK | 🇬🇧 UK Standards & Open to Logistics IT, Systems Support & DevOps roles

"If it runs reliably on a Core 2 Quad, it will fly on the Cloud. And if it survives a -25°C commercial freezer warehouse, it will survive enterprise production."

Before transitioning into full-time Linux systems administration, automation, and AI orchestration ("Vibe Coding"), I spent 7 years on the frontlines of UK distribution hubs (Avonmouth) operating Reach Trucks, voice-picking headsets, and RF handheld scanners, backed by 12 years of electrical component diagnostics.

In high-throughput warehousing, downtime is measured in lost pallets and missed departures:

• When a warehouse Wi-Fi roaming glitch drops an RF scanner mid-pick, pickers stop.
• When an industrial Zebra thermal printer jams or loses its driver spool, shipping docks stall.
• When a Warehouse Management System (WMS) drops a transaction socket, the whole conveyor line backs up.

I don't just write scripts; I build robust, fault-tolerant systems designed to withstand real-world operational chaos.

Active, battle-tested software designed to solve concrete operational and resource bottlenecks.

Desktop-Agnostic GPU VRAM & Local LLM Orchestrator

• The Problem: Running local LLMs (llama.cpp / llama-server on full GPU offload) consumes ~5.38 GB VRAM. Launching modern Steam games on a 6GB card (GTX 1060) triggers immediate Out-Of-Memory (OOM) crashes.
• The Solution: A universal systemd-aware wrapper that hooks into the Steam game lifecycle. It cleanly stops llama.service, flushes CUDA context (freeing 5.4 GB), enables NVIDIA PRIME offload, and launches the game with full VRAM. On exit, a GUI dialog (kdialog/zenity/yad) with a 60-second timeout automatically restores the AI service.
• Docs & Deep-Dive: English README | Česká dokumentace

Low-Hardware Real-Time STT & EN ➔ CS Translation Shield

• The Problem: Bridging real-time technical language barriers during live UK technical interviews and training without cloud latency or heavy dependencies.
• The Solution: A lightweight Python Textual TUI with a strict CPU-first fallback strategy and PipeWire/PulseAudio loopback routing. Features automated session logging in Markdown and single-command local deployment (~/.local/bin/myvoice).
• Docs & Deep-Dive: English README | Česká dokumentace

14-Container Docker MCP Gateway & Memory Backbone

• Architecture: Complete multi-agent MCP platform running 14 modular Docker microservices (Filesystem, Git, Terminal, Database, Memory, FORAI Code Tracking, Qdrant vector database, and Marketplace).
• The Hardware Irony: This core orchestration brain runs on the oldest, lowest-spec hardware in my entire lab. MCP tools and API gateways demand reliable RAM and network persistence, not raw compute cycles. Running the nervous system on legacy silicon proves zero-waste efficiency.
• Docs & Deep-Dive: English README | Česká dokumentace

Linux MSR CPU Frequency Control & Thermal Throttling

• The Problem: Upgraded a 2010 All-in-One PC with a significantly more powerful CPU, leading to severe thermal saturation and automatic shutdowns at 82°C.
• The Solution: Custom Linux power management suite interacting directly with CPU MSR (Model-Specific Registers) to dynamically step frequencies, control PWM fan curves, and multiplex core utilization to prevent thermal cutoff.
• Docs & Deep-Dive: English README | Česká dokumentace

Deep technical sandboxes where unfinished code yielded profound low-level systems knowledge.

• Deep-Dive Lesson: Pushing a resource-constrained 4-core ARM Raspberry Pi to deliver 1080p media playback and Steam Link game streaming.
• Kernel Orchestration:
  • Core 0: Isolated exclusively for network stack, Wi-Fi, and Bluetooth IRQ noise.
  • Core 3: Locked for ALSA audio stack to eliminate audio dropouts and buffer underruns.
  • Cores 1 & 2: Dedicated to the MPV player engine with dynamic 160% threshold burst scheduling to Core 3 during peak rendering.
• Docs & Deep-Dive: English README | Česká dokumentace

• Deep-Dive Lesson: Linux peripheral event automation via udev rules.
• Implementation: Triggers background library synchronization upon USB plug-in, leverages Gemini AI for metadata normalization, and reformats complex technical tables for readable E-Ink typography.
• Docs & Deep-Dive: English README | Česká dokumentace

• Deep-Dive Lesson: Documented post-mortem of 150+ configuration failures while managing multi-server SSH keys, port forwarding mismatches, and nested tmux sessions. Led to unified key management and automated environment provisioning.
• Docs & Deep-Dive: English README | Česká dokumentace | The SSH Chronicle

• Architectural Lesson: Originally designed a custom autonomous coding agent (MyCoder). When open-source pi (p-coding-agent) was released with modular provider support and zero bloat, I immediately deprecated MyCoder.
• Senior Takeaway: A pragmatic engineer never reinvents the wheel when an excellent open-source tool solves the problem.

Generated with automated GitOps discipline — Contact: GitHub Profile