One machine.
My old gaming PC, wiped and turned into a server with Proxmox. The graphics card that ran games now runs language models. You’re looking at it through a thermal camera.
16 logical cores31 GiB of memory1 GPU for local AI
DevOps engineer in Toronto
Harsh Kapoor
By day I keep AWS and bare-metal infrastructure running. At home, my old gaming PC is a server with fourteen apps and a local AI, and I run it like production too.
Two halves, fourteen apps.
Proxmox splits the machine. A virtual machine runs every app in Docker; a container gets the GPU and runs Ollama for anything that needs a model.
- Fuel
- Forgejo
- Grafana
- InfluxDB
- Uptime Kuma
- n8n
- Open WebUI
- AdGuard Home
- Jellyfin
- Syncthing
- ntfy
- Caddy
- Homepage
- Dockge
Run like production.
- Changes only through Git
- Self-hosted Forgejo, protected main, pull requests, mirrored to GitHub.
- HTTPS on every app
- Caddy with a Let’s Encrypt wildcard via Cloudflare DNS.
- Private by default
- Nothing is exposed; I reach it over Tailscale.
- Backups I’ve restored
- Nightly encrypted restic to Backblaze B2, restore tested.
- Watched
- Grafana with GPU telemetry, Uptime Kuma, alerts to my phone.
- Infrastructure as code
- Terraform creates machines on Proxmox from a file.
Things I’ve built
Runs on the homelab
Fuel
A calorie tracker I built for myself, with the AI running on my own GPU.
- Type “5 egg whites and 40 g oats” or snap a photo; a local model works out what and how much, and USDA data supplies the nutrition.
- Live barcode scanning in iPhone Safari through a WebAssembly reader.
- Targets that adapt: it estimates real energy burn from logged intake and the weight trend.
FastAPI, SQLite, JavaScript, Ollama, Docker. 31 automated tests.
Private repo, happy to walk you through it.
Observability
Watching the homelab
Dashboards for the Proxmox host, every VM and container, and the GPU, so I see problems before they become outages.
- GPU temperature, power and memory collected with Telegraf into InfluxDB.
- Proxmox metrics for each VM and container: CPU, memory, disk and network.
- Uptime Kuma checks every app, and alerts reach my phone through ntfy.
Grafana, InfluxDB, Telegraf, Uptime Kuma, ntfy.
Real screenshots from my Grafana. The GPU idles at 25 °C, the cool end of this page’s colour scale.
Network automation
NetDevOps Lab
A campus network defined entirely as code, built while studying for the CCNA.
- Two redundant core switches with OSPF and VRRP, VLANs for users and servers, on Arista cEOS in Containerlab.
- One inventory file is the source of truth: Jinja2 renders configs, NAPALM deploys them, pytest proves it works.
- Drift detection, plus a local LLM that answers questions about the live network.
Python, Nornir, Jinja2, NAPALM, pytest, Containerlab, GitHub Actions.
Local AI
AI Car Dealer
Reads a used-car listing and tells you whether it’s a deal, all offline.
- Scores the listing, flags scam signs and missing details, and drafts the first message to the seller.
- Typed output from a local model with Instructor and Pydantic, not loose text.
- Checks the VIN’s checksum offline to catch typos and fakes.
Python, Ollama with Gemma 3, Instructor, Pydantic, Crawl4AI, PostgreSQL, Docker Compose.
Research, 2023 to 2024
Observability for a 5G core
A cloud-native 5G core on Kubernetes, instrumented so you can see what it’s doing.
- Free5GC deployed on Kubernetes as a testbed.
- Prometheus and Grafana for metrics, with Python and shell automation for setup and data collection.
- Published as “Performance Monitoring of Cloud-Native 5G Core Network” (IEEE conference proceedings, 2025), with G. H. S. de Carvalho.
Kubernetes, Free5GC, Prometheus, Grafana, Python, Bash.
Also
- VastraA smart wardrobe app built with two friends: AI tagging from a photo, weather-aware outfit ideas, outfit sharing and virtual try-on. I worked on the build.
- ResumeXTeam résumé builder. I owned testing: 65+ Jest tests across payments and sign-up.
- RecipicaAndroid recipe app in Java and Retrofit with search and tag filters.
About me
I’m a DevOps engineer at Concentrix in Toronto: AWS and bare-metal infrastructure, CI/CD, infrastructure as code, and being on call when something breaks at 3 a.m.
Before that I spent two and a half years at Ontario’s Ministry of Transportation doing the same kind of work for provincial systems. The homelab is where I try things on my own terms, from GPU monitoring to running language models locally. I studied Computer Science at Brock University, and I’m working toward Kubernetes (CKAD) and the CCNA.
Experience
-
Oct 2024 to now
DevOps Engineer, Concentrix (via Wintab Tech Solutions)
- Keep highly available AWS environments and bare-metal servers running for a micro-service architecture, on a 24/7 on-call rotation.
- Automate with Python against AWS APIs, and Bash for Linux maintenance and OpenStack deployments.
- Run CI/CD with Jenkins and GitHub, with Terraform and Ansible for infrastructure as code.
- Set up observability with OpenTelemetry, Grafana, Splunk and PagerDuty to speed up incident response.
-
Mar 2022 to Sep 2024
DevOps Analyst and Software Engineer, Ontario Ministry of Transportation
- Orchestrated AWS micro-services and OpenStack deployments on bare metal for systems handling large provincial datasets.
- Wrote Python automation and telemetry tracing, and Bash for Docker provisioning and log parsing.
- Maintained CI/CD with Jenkins, Terraform and Ansible; triaged defects and validated releases against 24/7 health goals.
-
2020 to 2024
Honours BSc, Computer Science, Brock University
Research on a 5G core testbed, data structures, databases, machine learning and AI.
What I work with
- Cloud and infrastructure
- AWS, OpenStack, bare-metal Linux, Proxmox, Docker and Compose
- Automation and IaC
- Terraform, Ansible, Python, Bash
- CI/CD
- Jenkins, GitHub and GitHub Actions, Forgejo, SonarQube
- Observability
- OpenTelemetry, Grafana, Prometheus, Splunk, PagerDuty, InfluxDB
- Networking
- VLANs, OSPF, VRRP, Containerlab, NAPALM, Caddy, Tailscale
- AI
- Ollama, local LLMs, structured output with Pydantic
- Certificates
- DevOps Foundations, LinkedIn Learning (CompTIA-approved CEUs), 2026
- Preparing for
- Certified Kubernetes Application Developer (CKAD), CCNA
Let’s talk.
Want to talk DevOps, cloud or homelabs, or about a role? Send a message and I’ll get back to you.