HanzalaChaudry

Owners of busy restaurants do not reply to their Google reviews. Not because they don't care, but because it's forty minutes at eleven at night and it's the first thing that goes. So Titanium drafts the replies for them, in the voice of that particular business, and the owner reads them and sends.

Getting a model to write one convincing reply is an afternoon's work. Getting it to write a hundred that don't read as though a machine wrote them is a different problem, and it's the one worth explaining.

A reply goes through four passes. It gets drafted, then a second pass critiques that draft and rewrites it, because a model told to avoid a phrase will satisfy the ban lexically and leave the tell intact; you need something looking at the output rather than the instruction. Then it's trimmed to a word budget that shifts with the star rating, and finally checked against the rules that can't be broken. It is fenced off from inventing facts: it cannot name a dish the reviewer never mentioned.

The part I'd point at, though, is the ledger. Google removes owner responses that mirror each other across a profile, and a model has no memory of what it wrote yesterday. So every reply that actually goes out is recorded, and the last thirty are mined for repeated openings, closings and three-word runs, which are then handed to the next draft as a list of things it may not say. Variation is enforced outside the model, because it cannot be trusted to the model.

A house in Leyton ran its lighting on Rako, a proprietary control system. Thirty-odd circuits, every light in the building behind it. Then it died.

Nobody would take the repair on. A dead proprietary system with no surviving documentation is awkward, unglamorous work, and the safe quote is always to rip it out and start over. So it sat. The owner, an elderly man, lived off the emergency lighting instead. For years.

I kept the wiring and replaced the brain. Every dead Rako module came out and a Zigbee relay went in on the same circuit, each one labelled with the room it actually controls: hallway chandelier, living room bay, kitchen pendants. The original paperwork was long gone and I wasn't going to leave the next person guessing. A Raspberry Pi running Home Assistant took over as the controller. I put in gigabit Wi-Fi so the mesh had something solid underneath it, then built the media wall in the living room and brought that onto the same system.

Working lights were the floor, not the job. A switch on a wall is still something a person has to think about, and this owner is elderly, which is most of the reason the dead system went unmentioned for as long as it did. So the house got told how to decide.

The lighting runs off sun elevation and the local cloud cover rather than a clock, so an overcast afternoon in February and a bright evening in June get different answers at the same time of day, and the house comes up gradually rather than snapping on. The blinds work to the same logic: down against low sun on the front elevation, open when the room would otherwise want a lamp. A camera on the driveway reads number plates, so the gate opens for the cars that live there and stays shut for everything else, which matters more than it sounds when opening it used to mean walking out to it.

None of that is a smart bulb in an app. Nobody in the building operates anything. It makes its own decisions off its own inputs, and the only time the owner touches it is when he wants something other than what it chose.

Titanium is a review platform for hospitality that runs on hardware I make. Someone taps a station on the counter. If they're happy they go to Google; if they're not they reach the owner privately while they're still in the building, with the Google link on screen either way. Behind it sits a dashboard for every site: league tables, complaint themes, the drafted replies above, and a weekly report an owner can read in two minutes.

I built all of it on my own. The stations are modelled and printed here, per client, with their branding and no supplier logo on them. So are the platform, the AI, the payments, the contracts and the selling.

A dated conservatory on the back of a semi, used mostly as somewhere to put things, came off. A proper extension went on in its place: foundations dug, cavity walls up, the back of the house opened out into one room.

Brickwork is the straightforward part. What the job actually turns on is the set of decisions before it. Where the island sits settles where everyone ends up standing. The lighting has to work at seven in the morning and at nine at night, which is two schemes rather than one. And a herringbone floor either reads expensive or reads like a mistake, depending entirely on how the run meets the units.

Packing delivery orders was slower than it needed to be because the label rolls tangled and fed badly. Nobody had raised it. That was just how the station worked. I modelled a mount that carries the roll on a bearing so it pays out flat, printed it, and put it on the pass.