Derek Gee

One builder.
A small fleet of software.

Product manager by day. After hours: self-hosted apps for one household, built alongside AI agents that plan, review, and ship.

LocationSF Bay Area
Day jobPrincipal PM · Financial Reporting
BackgroundTax → Analytics → PM
Vaults4
Support systems3
Ships fromlaptop + phone

The Fleet

one vault per corner of the household

Support systems

not vaults — the projects that keep the fleet running
cC In service

claude-config

The shared playbook for working with AI agents: working agreements, skills, and guardrails, versioned in git and synced to every machine — so each agent session starts with the same context I would give a new teammate.

Roleagent playbook
Reachevery machine · every repo
hA In service

homeAssistant

The physical side of the household: Home Assistant automations for the rooms the vaults only keep records about. Same private network, same philosophy — the house's data stays in the house.

Rolehome automation
Hostdedicated box · private network
nM Active build

networkMonitor

Keeping an eye on the private network itself: which devices are on it, what's up, what's down, and who should be told.

Rolenetwork visibility
Stageactive build

The System

machines, agents, and how work flows
Laptop + phone
where I work — a surprising share of commits start as SSH sessions from a phone
Dev machine
always-on desktop where coding sessions and builds live
Agent sandbox
a dedicated, isolated machine runs an always-on assistant on OpenClaw
Home server
git origin + Docker host; where the vaults actually run
Private mesh networknothing exposed to the public internet

Agents plan tasks, write and review pull requests, and handle routine operations. Anything that touches production waits for a human sign-off. The sandbox machine exists so agent work is isolated by construction, not by promise.

The Operator

who runs this thing

Principal product manager, working in financial reporting, in the San Francisco Bay Area. Before product: a decade in tax accounting, then business analytics — which explains both the daily net-worth habit and the tolerance for well-kept records.

Building software for my own household is the craft I practice after hours. Strong in Python, increasingly at home in TypeScript, and opinionated about clean branches, Conventional Commits, and tested code — because the agents hold me to it.