Things I built because nothing off the shelf fit.

About thirty of them. Most began as something I was already paying for, or something that didn’t exist at all.

I buy technology for a living. These are what I do with the other half of the idea — the half where you find out whether the thing actually works.

Try these

MyPetLife

An all-round pet management app for iOS and Android. Pet profiles, vet and vaccination tracking with booster reminders, medications with dosing schedules, feeding logs, weight, and a photo journal — for a dog, a cat, or a rabbit, all in one place.

Try it in a browser → mypetlife.savvaskonstantinidis.gr — no install.

Or on your iPhone: My_Pet_Life on the App Store — free.

BookMe

A scheduling and booking app I built and self-host it. It started as a Calendly replacement; it turned into the tool my flat viewings actually run on.

• A public booking page, no account needed
• Double-booking prevented in the database, not the interface — the guarantee holds even when two people click at once
• Busy-time rules that read my own calendars, so it never offers a slot I can’t take
• Email confirmations and reminders, sent from my own server

→ Open the booking page

GatherApp

A social planning platform. Getting a group of people to agree on a time and a place is a problem software keeps failing at, so I had a go.

Plan activities with friends — without the group chat chaos.

→ Open GatherApp — live, in a browser, no install.

Why I build them

Three reasons, and they’re not equally noble.

• The tool didn’t exist. Nothing does what I wanted for keeping a record of everything in the house, or for logging a motorcycle ride properly, or for a family of five’s kitchen.
• I was renting something I could own. A subscription that holds my data, breaks when the company pivots, and charges me monthly for the privilege. Several of these replaced exactly that.
• I wanted to understand the technology I buy for a living. This is the honest one. You cannot sit across a table and assess a cloud agreement, a consumption-based AI contract or a vendor’s lock-in claims if you have never run the thing yourself. Building is how buying gets better.

What they replaced

This is the part I’d put on a CV if CVs had room for it.

• BookMe replaced Calendly — a monthly fee per seat for a booking page, and my calendar data sitting on someone else’s server.
• ZeBill replaced a budgeting subscription, built from the parts of YNAB, BlueCoins and OutBudget that were actually worth having.
• Keepr replaced a chain of attempts — Notion lists, Obsidian lists, HomeBox, Itemtopia — every one of which I outgrew or abandoned.
• Family Hub replaced a family chat assistant with something the whole house can actually use and see.
• A Microsoft 365 subscription was cancelled outright as part of the same instinct: stop renting, start owning.
• And the de-Google work — mail and services moved off the ecosystem entirely.

The pattern in all of them: my data, on my hardware, on terms I can read.

How they get built

Not the way you’d expect. Almost none of this starts with opening an editor.

The process

• A spec interview first, always. Before any design, I sit with my AI and we work out what the thing actually is. No drafting, no mockups, no code until the questions are answered. An unanswered question is not consent to build — it’s a reason to wait.
• Every project starts as a note and a repository, on the same day, before it has a single line of code. If it isn’t tracked it isn’t real.
• A working visual before a build. Anything with an interface gets a mockup first, in the real colours and the real typefaces, so the design is settled before it’s expensive to change.
• Build, then test, then audit. My AI builds it. I test it as a user — which is where the real defects come from. Then an independent audit pass looks at it, deliberately, as someone trying to break it.
• An honest empty state beats a flattering one. If a panel has no data, it says so. Nothing gets mocked up with invented numbers, because an invented number in a mockup becomes a claim on a live screen.

The principles

• Verify, don’t assume. Test all 215 URLs, not a sample. Read a write back before believing it landed. Twice this month a mistake of mine survived a check that only looked convincing.
• Dry-run anything irreversible. Especially anything that publishes, sends, or deletes.
• Nothing gets deleted without asking. Not a file, not a config, not a page.
• Write the lesson down where the work happens — the pitfall goes into the procedure, so the second time costs nothing.
• Show your work, including the mistakes. The errors are usually the interesting part.

The tools

Astro · Cloudflare Workers · Fastify and Prisma · PostgreSQL · Tailwind · Docker · Home Assistant · Gitea · Claude Code and Codex for heavy code lifts · and a self-hosted Hermes agent doing the orchestration, the research and the bookkeeping.

Working with a machine

Every one of these is built by a hybrid team of one human and a small fleet of AI agents, and the division of labour is the interesting part.

• I decide. What to build, what matters, what the trade-offs are, what ships. Strategy, judgement, and taste don’t delegate.
• It executes. Research, drafting, code, configuration, testing, verification, and the tedious discipline of checking every URL rather than three of them.
• And it argues with me. The most valuable thing I’ve built into it is permission to disagree — to say a plan is wrong, name the risk, and propose something better. An assistant that only agrees is an expensive echo.

It isn’t that the AI writes the software. It’s that the distance between having an idea and having a working thing has collapsed.

What they cost

No licences. No per-seat fees. No subscription that renews whether I use it or not. The bill is my time — and this is the honest accounting of it, read from my own time log for 2026.

Project	Hours in 2026	What it is
ZeBill	99.7	Personal finance PWA
Family Hub	88.9	Family chores, calendar and house guide
BookMe	38.5	Scheduling and booking
Agent Cockpit	20.4	Dashboard for the AI agents
QikStart	19.5	Motorcycle ride recorder (Android)
Travel League	17.0	Friends’ travel log and XP game
Pledge	16.5	Social commitments
Micro:bit display	16.2	Physical status display for the homelab
Keepr	16.1	Home inventory
Lifebalanc.er	12.9	Life balance tracker
Life Performance Dashboard	12.8	OKRs across nine life areas
Project Portfolio Dashboard	11.2	Everything in flight, at a glance
MyPetLife	7.2	Pet management app
GatherApp	2.0	Social planning

Call it 380 hours across just those — a solid nine weeks of full-time work. Done in the margins of a demanding job, mostly between nine and midnight.

What I can’t give you is a cost per app, and I’d rather say so than estimate one. The AI bill is real but it isn’t itemised per project, so any per-app figure would be invented. What I can tell you is the shape of it: a fixed monthly cost that doesn’t grow when I add a project, against software that would have been €50–150 a month in subscriptions if I’d bought equivalents of all of it.

The rest of the shelf

Built and in daily use

• Family Hub — chores with points and a leaderboard, an AI chat to add tasks and send reminders, a calendar snapshot, and a house troubleshooting guide for the whole family.
• ZeBill — offline-capable personal finance: envelope budgeting, goals, multi-currency, five-year forecasting.
• Keepr — every thing, where it lives, and its warranty. 127 items catalogued.
• TapTally — a multi-counter clicker. Trivial, and used constantly.
• FeelTally — a sentiment log for tracking mood over time.
• Project Portfolio Dashboard — every project and its next action, from anywhere.
• MealBrain — recipes, pantry stock, shopping list and meal plans for a family of five.
• Micro:bit status display — a physical screen and sounder on my desk that tells me what the homelab is doing.
• Agent Cockpit — the control room for the AI fleet: what each agent is doing, what it has done, and what it has cost.

In progress

• Life Performance Dashboard — real-time progress against my own objectives across nine areas, fed by journaling, Garmin, finance data, energy monitoring and a bedside sleep tag.
• Travel League — friends log where they’ve been and earn XP. Leaderboard, levels, and a map of places more than one of us has reached.
• QikStart — motorcycle rides recorded from the phone’s own sensors: distance, speed, stops, a smoothness score, and fuel history.
• Pledge — commitments made to other people, kept visible.
• Lifebalanc.er — where the time actually goes.
• Explorra — a personality-driven travel platform: a curated destination knowledgebase with an AI that designs itineraries from it.

On the bench

• HUXE Clone — a personalised morning audio briefing: news and follow-ups, in my own voice.
• The Bookie — a prediction betting game for family and friends. Play money, real bragging rights.
• EcoSim — an ecosystem simulator built with my daughter: real ecological modelling, real species parameters, wildfire and drought events you can inject.
• HΩM — a small autonomous rover for the house, built with my son. He does the machine learning; I do the hardware.
• Tai Chi Visual Instructor — a practice programme that plays with timings, cues and an animated demonstration.
• Spotify DJ — a DJ front end for the house speakers, which Spotify’s own API can’t see.
• MyFamilyCare — one shared space for a parent’s care: documents, medications, appointments, handoff.
• My Shopping Butler — the shopping list, handled properly.

Some of these will ship. Some are already teaching me something. A few I will quietly abandon, and that’s fine — they cost me an evening each, not a company.