NICKHINDS

You type a request like “spin up two Ubuntu servers, 8 GB each,” and the AI writes a plan. My code checks it against the live server, asks for approval, takes a checkpoint, and only then runs it. The AI never gets a direct path to the server.

Most of the design went into controlling what the AI can do. Every action passes one checkpoint, destructive actions start off, and the AI can be denied things I'm still allowed to do.

Python · FastAPI · Proxmox VE · Claude · Telegram · systemd

A training analytics platform that pulls ride and wellness data straight from Garmin, archives the raw files immutably with checksums, and stores everything in a database I control. It computes the metrics a coach would actually look at, normalized power, TSS, the CTL/ATL/TSB load model, power curves, and aerobic decoupling.

The part I like most is the MCP server on top of it. I can ask Claude real questions about my training in plain language and get answers grounded in my own numbers, because the model queries the data store instead of guessing.

Python · SQLite · MCP · Garmin

Cameras that turn off when I'm home and on when I'm away, silently. iOS Shortcuts couldn't do it without notification spam, false triggers on every WiFi blip, and a hard dependency on my phone staying connected, so I moved the whole thing onto a server that runs all the time.

Presence detection runs at the network layer over ARP, roughly thirty milliseconds instead of waiting on a phone-side check-in, with a state machine and grace periods so a brief dropout doesn't flip anything. It runs as its own unprivileged system user with credentials encrypted at rest, and it stays on the local network by design.

Python · Flask · systemd · ARP detection

An engine that scans a network, keeps an inventory of the assets and services it finds, and raises events when something opens, closes, or repeats. Every event carries a severity and the exact observation behind it.

It runs from the command line, from Telegram, and now from an operator console, and all three call the same functions underneath. The AI can narrate a report or, under a grant that expires on its own, trigger a scan, but it never touches severity.

Python · SQLite · systemd · Telegram · CLI

A single script that stands up a private DeepSeek model with Docker, Ollama, and OpenWebUI. Simple enough that anyone could run their own local AI without touching the command line twice.

A menu-driven installer that automates homelab setup across security monitoring, system administration, networking, and application delivery. The goal was making the whole stack approachable for someone starting out.

A centralized network boot server on iVentoy and Ubuntu, booting machines over the network and managing ISO images from a single place instead of chasing USB drives.

A credentialed Nessus patch audit across Windows, Linux, and macOS, with a remediation plan that prioritized the high-severity findings first.

A free, open-source SIEM with automated agent deployment, centralized log aggregation, threat detection, and CIS benchmark compliance checks across the lab.

A containerized OpenVAS deployment running both credentialed and uncredentialed vulnerability scans against the home network, then working through what it surfaced.

A credentialed Nessus scan against a MacBook through a dedicated service account, remediating the high-severity vulnerabilities it found.

Hardened a client WordPress site end to end: a staging environment for safe testing, MFA, role-based access, TLS, and disciplined update management.

A firewall appliance built from scratch on pfSense with custom rules, IDS/IPS, DNS sinkholing, live traffic analysis, and network segmentation. This is where the OSI model stopped being theory.

Migrated homelab services off licensed Windows VMs onto lightweight Linux, cutting cost and improving stability and uptime while learning my way around the command line properly.

Designed and simulated a segmented SOHO network, then studied how traffic moved through it and ran controlled attacks with Wireshark, Nmap, and Kali on an isolated lab segment.

Deployed an OpenVPN server on Linode, a cleaner path than the Raspberry Pi and local VM builds I'd done before it, and my first real use of cloud infrastructure as a service.

Virtualized the entire homelab on Proxmox, consolidating several physical machines into one manageable, scalable server. The foundation everything after it ran on.

Long before any of this was written down, it was taking computers apart, breaking them, and fixing them again. Everything above is just that same instinct, documented and pointed at bigger problems.