Why · the short story

Just take pictures.

We take pictures of everything. Restaurants, obviously. Wine, obviously. But also the plate at a lunch that just looked good, the street, the door, the sign. Mundane or important, it goes on the roll, and you decide later which it was.

We tried the apps. The wine one, the food one, the one that wanted to know about the shoes. Every one of them starts empty and asks you to fill it — at dinner, with the bottle in front of you and people waiting. That’s friction, and friction wins. You keep it up for a week, maybe a month, then the habit dies and the app dies with it.

But the pictures kept coming. Fifteen years of them — everything since the first iPhone with a decent camera; about 28,000 on my roll alone. The record already existed; nobody had read it. Nobody needed a new habit. The roll needed someone to read what was already shot.

I run Vertesia, and I’m an engineer. What we do for our customers, over and over, is help them make sense of their content: read it, structure it, turn it into an ontology they can actually use. My camera roll was the same problem, at home — fifteen years of content nobody had read. A personal problem that, I think, everybody has. I was like: why not.

That’s what Balise does. It reads the camera roll you already have and turns it into collections you can ask. Every bottle you ever photographed becomes a cellar. Every look, a closet. Wrists, tables, places, the fish. Not a wine app plus a watch app plus a closet app, each starting empty. One place, after the fact.

So I iterated a bit, pointed it at my own roll and ran the import. The measured engineer’s reaction I would like to report here is not the one I had. What I said was: holy shit.

There they were. The watches. The wines. The looks — even the odd, funny stuff. Not a folder of pictures of watches: the watches, with their references and the days they were worn. Nobody typed any of it. It was awesome.

And not just me. We ran it as an experiment on other people’s rolls — same app, nothing configured — and everybody had some version of the same reaction: oh, awesome. Now I can use that thing. Every roll came back a different library: the looks and the labels from one, the wines and the restaurants from another. Nobody had told it what to care about. It read that off the pictures — which is the whole trick, and the reason this was never going to be a wine app.

And once the base is there, you stop scrolling and start asking. Which watch was that in Buenos Aires. Where was that Puligny. What was the name of the place in Geneva with the terrace. The agent goes through what was read off the pictures and comes back with the answer, not a grid of maybes.

Why “Balise”

Balise is French. It’s the marker you leave so you can find the place again — the two painted bars on a rock on the GR trails, white over red, that say: this way. The buoy over the channel. A balise is a beacon. Pictures are the beacons of your life; that’s why you take them. We keep the marks.

What we built

An iPhone app. You keep the normal Camera. You open Balise, pick what it may see, and tap Import. It reads a small copy of each photo, works out what it was, and files it: the bottle and the vintage, the watch and the reference, the restaurant, the city, the date. Screenshots and blur, it skips. Then you browse, or ask in plain words. “The Burgundy from Lyon.” “Which lake was the trout.” Apple finds the photo. Balise keeps what it meant.

When it gets something wrong, you fix it in one tap and add what you remember. It gets better because you talk to it, and you never changed how you shoot.

Point it at a big roll and the first stretch is askable in about five minutes. The rest lands over the next hours. Then it keeps up: on a plan, what you shoot from that day on comes in on its own. The roll stays the way you already use it, and the collections grow behind it.

Where this goes

Photos are probably what we use most on our phones. We capture everything, all the time, and almost none of it works for us. AI can finally read it. That’s the whole idea: take what already exists and hand it back to the person who made it.

The order never changes. It looks at what you already shot. It builds the base. Then it answers — which night, which watch, what to order, what to pack, where you’d actually like to stay. Not what a review site recommends to everyone: what fits the person who took these particular photographs of these particular rooms and plates and streets, and clearly liked them, because they stopped and took the picture.

What we refuse to build

A second camera. If you ever open Balise to take a picture, we failed.

A face yearbook. No people graph, no who-knows-whom, and no facial recognition running on our side. The only face work happens on your phone, in Apple’s own frameworks, and the only person it can learn is you — only if you say yes. Everyone else in your pictures stays a stranger to us.

Training on your pictures. We don’t. Our providers don’t either.

Keeping your pictures. We make data, not copies. Your originals stay in Apple Photos; the small copy that goes out to be read is deleted, and there is no archive of your images on our side. Hidden and shared albums stay out in the first place.

Quietly doing something else with it. What Balise reads is private today, on every plan: not sold, not shared, not fed to anybody else’s product. If a free tier ever has to pay for itself another way, that will be a choice put in front of you in plain words, with a switch — not a clause that appears in an updated policy one Tuesday.

That’s how we’d want it built if it were our roll. It was.

Who it’s for

People who already take the picture. The one who shoots every label and never opens the app meant to log it. The one whose roll is half outfits. You’ve been building the collection for years, mundane and important in the same stream. You just never got to see it.

The ask

Point it at a roll you already shot. One album, or one trip. See what comes back. Then tell us what it got wrong; that’s the part we want most.