· 6 min read · mpoapostolis
874 Commits: What Building Galerra Actually Looked Like
Fifteen months, two codebases, 874 commits, one person. What got built, what got thrown away, and the weeks where nothing moved — read back from the record rather than from memory.
Galerra turns photographs of your work into a walkable 3D exhibition with one link. I have been building it alone since June 2025. There are 874 commits across fifteen months and two separate codebases, and the record is a more honest account than my memory — it shows the weeks I shipped nothing, and the things I deleted, as plainly as the things that survived.
June 2025: eight days to something that worked
The first commit is called, unhelpfully, “initial”. By the end of that same day there was a profile page, artwork upload, and somewhere to put it.
Eight days later the whole loop worked: upload a photograph, get a wall, walk up to it, send someone the link. That fortnight was the fastest the project has ever moved — seventeen, eighteen, twenty-two commits in a single day — and it produced a version of Galerra you would recognise immediately and would not want to use for long.
It only had a handful of rooms. They were fixed 3D models, and the position of every painting on every wall had been typed in by hand, one at a time. It worked beautifully for a show of exactly the right size and fell apart for anything else.
I also rebuilt the way images load about five times in one week. Getting a photograph onto a wall in 3D sounds like the simple part. It is not.
July 2025: open, and then over
On 5 July it opened to anyone who wanted to try it. Artists walked their own work for the first time, and told me what was wrong with it.
Looking back through that month, the pattern is clear and it is not flattering. From 20 June onward I had stopped adding anything and started rearranging what was already there — a new explore page, a new landing page, a new profile layout. On two consecutive days at the end of June there are six commits in a row rewriting the same paragraph on the homepage.
I was rewriting sentences because the thing underneath them could not do what I wanted, and rewriting sentences is easier than admitting that.
The last commit of the first Galerra is dated 17 July 2025. Then ten weeks of nothing.
September 2025: starting over
I did not repair it. On 27 September a new, empty project opens, and the 3D side was replaced before a single feature was written.
That was the decision the ten quiet weeks produced: the rooms could not be fixed models with hand-placed paintings. They had to be built around whatever work you actually have — five pieces or forty, tall or wide, one room or six. Everything you use today grows from that, and almost nothing was carried across.
October was one hard fortnight — most of that month’s work happened inside a single week. Then it stopped again. November 2025 has two commits in the entire month.
I would like to tell you there was a strategy behind the pause. There wasn’t one.
Winter 2025–26: building things in order to delete them
In December I made galleries multiplayer. You could see other visitors moving around the room with you, and talk to them.
In February I took it out again. It was demanding to keep working, almost nobody used it, and — this is the part I only understood by watching people — a gallery you walk alone is closer to the real thing. You slow down when nobody is watching.
That winter also rebuilt the look of everything: warmer light, cleaner walls, frames that sit properly, rooms that open into one another instead of stopping.
Other things built and then removed, in order: per-gallery subdomains, gone the same day they shipped. An entire advertising tracking stack, seven commits undone by one. Private galleries, removed in June 2026 — a “private” gallery only its owner could see was not a feature, it was a promise the product did not keep.
I think the deletions say more about the product than the additions do.
March 2026: the biggest month
161 commits, the largest of the whole history, and the 3D viewer was barely touched in any of them.
The last weekend of March was a hundred commits over two days. Saturday built analytics — so an artist can see how many people walked their show, and which piece they stopped at. Sunday built the social layer: following an artist, saving a gallery, leaving a note.
On the Sunday I also made the flat list of artworks the default way into a gallery, and reverted it back to the 3D room a few hours later. Both versions are defensible. The room won because the room is the point.
Something quieter started that month too: a classroom page, hidden behind a flag, that twelve days later became a real thing teachers could use.
Summer 2026: the catalogue, and five silent weeks
July was the best month of building in the project’s life. Seventeen commits in one day on how a room looks — floors, wall paint, the way a frame meets a wall. Then the exhibition catalogue: every gallery also opens as a book you flip through, with a printed plate for each work.
After 21 July there is exactly one commit until 5 September. Five weeks of nothing, straight after the busiest stretch of the year.
That is the rhythm the record shows, over and over: a hard push, then a stop. I have stopped pretending it is a flaw in my discipline. It seems to be how the thing gets made.
September 2026: the door
The restart in September began with a document instead of a feature — an honest look at what was actually broken.
The biggest thing on the list was the classroom. Nineteen teachers had set one up over five months. Eighteen of them got zero students in.
That was not the market saying no. It was my own front door: it asked a fourteen-year-old for an email address and a password, which means asking a school for permission, which never arrives. So a student now types a name and starts uploading. Nothing to approve, nothing to provision, no personal data collected from a child.
It took a day and a half to build. It had been sitting there, wrong, for five months, because I never went and stood at my own front door.
Fifteen months, two codebases, and a lot of people who have made something here. If there is a lesson in the record it is not “keep going” — it is that the things worth keeping were usually found by throwing something else away.
The current version is at galerra.art.