Quadro RTX 4000
GPU 1 81:00.0- Architecture
- Turing, 2,304 CUDA cores, 288 Tensor cores
- Memory
- 8 GB GDDR6
- Power cap
- 125 W
- Runs
- Ollama local language models (gemma4:e4b)
8 of 12 GB pooled VRAM66%
root:~$ whoami
Technical Support Engineer, Cloud & DevOps.
I design, secure and run a production-style private cloud on one Linux server, and I build the software that runs on it. It hosts 28 services for media, messaging, automation, smart home and local AI, published through Cloudflare with an origin that only answers Cloudflare.
$ whichClick a technology to see where it runs on my server.
Select a layer to see what it does and how it is configured.
Pick a scenario and press Play to follow the packet hop by hop through the stack.
Both NVIDIA cards are passed into containers through the NVIDIA container runtime. They run language models, photo recognition, vehicle detection and video transcoding on my own hardware, so prompts, photos and camera feeds never leave the house.
Prompts, photos, face data and camera feeds are processed on my hardware. The model API sits on an internal Docker network and is never published.
Hard caps keep AI from starving everything else: the LLM runtime is limited to 24 CPUs / 32 GiB RAM, photo ML to 12 CPUs / 24 GiB, and URXFilms runs critiques one photo at a time in its own GPU lane.
Workstation GPUs with 50 to 125 W caps give me article writing, photo critique, smart photo search and transcoding with no per-token or per-seat fees.
I write these myself and run them on the same server, with staging, snapshots and rollback.
Self-hosted n8n with its own Postgres and Valkey runs publishing, AI writing, sensor logging and chat bots on schedules. On my Windows machine, PowerShell scripts handle the daily chores around my projects.
An ESP32 air-quality node I programmed feeds a 7-inch touch panel I built with ESPHome, Home Assistant and my n8n reports.
Sample reading from the node's /json endpoint, 24 Oct 2026 01:03. BSEC accuracy 3/3 (fully calibrated).
Each service is described by what it does. Turn on software names to see the tool behind it.
Bulk data lives on spinning disks and anything latency-sensitive lives on NVMe. Every app gets its own dataset, so quotas and snapshots apply per service. Select a pool to inspect its datasets.
Before any schema migration touches production, I clone the live dataset and run the upgrade against the clone first.
Every container has a restart policy. 32 running containers define healthchecks, and all 32 reported healthy at audit.
About 38 Docker bridge networks from a custom 172.16.0.0/16 pool, one per stack. Databases publish no host ports.
CPU and RAM caps on every catalog app. LXC guests are pinned to CPU sets with memory caps.
no-new-privileges, targeted capabilities instead of privileged mode, unprivileged LXC for nested Docker, non-root images.
Snapshot, tag the running image, build, redeploy, check health and logs. Rollback is one retag and one rollback command.
n8n reports to Telegram and Discord; CrowdSec decisions sync to Cloudflare; Prometheus scrapes the photo platform every 5 s.
| Schedule (click row) | Automated job |
|---|---|
| every 1 min | Sync camera SFTP logs to the IDS (inode-preserving rsync) |
| every 2 min | Push CrowdSec ban decisions to a Cloudflare IP list (my Python script) |
| every 5 min | Sync reverse-proxy access and error logs to the IDS |
| every 10 min | Log the bedroom air-quality sensor to n8n |
| every 2–3 h | AI news scout writes a draft article; publisher posts the next one to WordPress |
| 8× a day per page | Auto-posting to 7 social media pages with random delays |
| daily | Copy the renewed TLS certificate into the encrypted-DNS server; 10:00 air-quality report |
| weekly | Refresh Cloudflare IP ranges with validation and rollback; recursive ZFS snapshots |
| monthly | Hash the archive to find duplicate files |
My DevOps course took one small Node.js app through every stage of a delivery pipeline, and I completed all of it. Select a stage to see the exercises behind it, or run the whole pipeline.
Four courses by Amazon Web Services that take the same ideas to the cloud: AWS foundations, then code, build and test, release and deploy, and operate and monitor. Select a course, or run a deployment through all four.