My technology stack is a continuously evolving operational philosophy. I maintain a Linux shop that prioritizes stability, security, and the ability to deploy simple, scalable stacks on single and clustered VPS instances. My infrastructure is designed for performance, maintainability, and predictability.
Operating Systems and Foundations
I have standardized my environment to prioritize long-term stability and consistent package management.
Primary Server Environment: Ubuntu 24.04 LTS (Noble Numbat) remains on most production nodes, with the fleet now staging its move to Ubuntu 26.04 LTS (Resolute Raccoon), which reached final release on April 23, 2026 (Canonical, 2026). Direct upgrades from 24.04 open with the 26.04.1 point release, due August 6, 2026, which is when the rolling cutover begins in earnest.
Primary Desktop Environment: Linux Mint 22.3 “Zena” (January 2026). My daily driver for development, writing, and administrative tasks.
Specialized Environments:
- FreeBSD 14.x: Utilized for high-reliability, network-focused server tasks.
- Windows 11: Maintained via Virtual Machine for specific compatibility and software dependencies.
Infrastructure, Hosting, and Deployment
I favor dense, efficient hardware that minimizes overhead while maximizing compute.
Deployment Management: Coolify (v4.1.x stable) acts as my primary self-hosted PaaS, managing the lifecycle and orchestration of my applications. The v5 rewrite, centered on native multi-server scalability, is now in active development, but Coolify has not committed to a ship date, so v4 stays the production target for the foreseeable future (Bacsai, 2026).
Hosting and Compute:
- Services: Hostinger is the primary provider for my service layers.
- VPS Compute: OVH is utilized for specialized compute nodes.
Accelerated Hardware:
- Creative and Prototyping (Mac Studio M2 and M4 Mini): Accelerated via Apple Metal and MLX.
- Production Inference and Bioinformatics (Linux Fleet): Accelerated via NVIDIA CUDA Toolkit. This is the backbone of the Digital Workroom logic, handling model fine-tuning and genomics simulations.
Virtualization: Proxmox VE 9.2 is utilized for containerization and virtual machine management across a 45-node lab network. The 9.x line brought a Debian 13 “Trixie” base, kernel 7.0, and a dynamic load balancer for the Cluster Resource Scheduler, which has meaningfully reduced manual intervention on node rebalancing (Proxmox Server Solutions, 2026).
Connectivity and Networking
Networking:
- Tailscale: The mesh fabric of my network. It provides secure, zero-config bridging for all remote Linux nodes.
- PIA (Private Internet Access): Utilized for VPN requirements and secure internet egress.
Storage: We utilize Asustor NAS solutions for local redundancy and off-site synchronization.
Data and Logic
My stack focuses on high-uptime, low-friction utility for data processing and software development.
Databases: SQLite, PostgreSQL, Supabase, MySQL.
Languages: Python, R, C++, PHP, Bash.
The AI Ecosystem (HarnessAI)
My workflow is automated through the HarnessAI framework, utilizing both proprietary frontier models and a robust suite of local, open-source models for specialized tasks.
Frontier Models (Cloud-Based)
| Model | Version / Notes |
|---|---|
| Claude Opus | 4.8 (current flagship, Anthropic) |
| Claude Sonnet | 5 |
| Gemini | 3.1 Pro (still Google’s shipping flagship; 3.5 Pro remains in partner testing with no public release date as of late July 2026) |
| GPT | 5.6 Sol (released July 9, 2026, replacing the retired 5.4 line; Terra and Luna variants available) |
| Perplexity | Current |
| Midjourney | Current |
| Suno | v5.5 (March 2026) |
Open-Source Models (Local Inference)
Deployed via Ollama across the Linux cluster and the Mac M2 Studio.
| Model | Parameter Size |
|---|---|
| gemma4:31b (dense) | 31B |
| gemma4:26b (MoE, 4B active) | 26B |
| gemma3:27b | 27B |
| gemma3:4b | 4B |
| qwen3.6 (latest Qwen generation) | Multi-size |
| qwen3:8b | 8B |
| qwen2.5-coder:32b | 32B |
| llama3.3:70b | 70B |
| llama4:scout | 109B |
| deepseek-r1:32b | 32B (distilled) |
| phi4:latest | 14B |
| gpt-oss:20b | 20B |
| cogito:14b | 14B |
| devstral-small-2:24b | 24B |
Cloud-only (through Ollama Turbo): kimi-k3, minimax-m2.5, glm-5.