In 2025 I spent a year building my own product. It did not become a business. I lost about 6,000 €, gave up a year of salary, and came back to a job. I would do it again, and this is why.
This is the story in the order it happened, including the part I never wrote down at the time.
An idea that was not mine
I was not looking to start a company. The idea came from friends: tradespeople in France are buried in admin, have no time for their own visibility, and pay too much for low-quality lead platforms. “There is no Doctolib for plumbers”, one of them said.
At first it did not interest me. Too broad, too far from what I knew. Then, late 2024, LLMs got good enough that I could picture a system handling the boring parts for a tradesperson without asking them to change how they work. The technical problem hooked me. I had just finished a freelance contract. I had savings. I said yes.
A friend came in as co-founder to carry the business side. We did fifteen interviews with carpenters, plumbers and renovation contractors, wrote down what they complained about, and found the same three things every time: paperwork, visibility, expensive leads. On the slides, Marty solved all three. In practice, one person could not build that in a few months, so I cut it down to visibility: a profile, a search engine, a way for clients to find a good tradesperson nearby.
Building it was the easy part
I had never built a product from zero. Data pipelines, ML platforms, backend APIs, yes. A whole product with a signup flow, payments, images, a search engine, a blog for SEO, a mobile layout, no.
I kept the stack boring on purpose: Next.js and TypeScript for the front, Python and FastAPI for the back, PostgreSQL with PostGIS and pgvector, Docker on GCP, Terraform for the infrastructure. Nothing I would have to learn while also learning everything else.
For five months I was the product manager, the designer, the front and back developer, the ops, and the QA. One day I was tuning a geographic search, the next I was moving a button because a plumber could not find it on his phone. I had planned three months. It took five. The product was finished in early June.
It was the most fun I have had writing software, and I am proud of what came out. That is not the point of this article.
The empty room
My co-founder had left the project a couple of months earlier. He had carried the commercial side and the sector knowledge. I do not hold it against him, and this is not the place to explain why he left. What matters is that when the product was done, I was alone in a room with something to sell, and nobody to sell it.
So I did what I knew. I wrote content, tuned pages, worked the SEO hard. Tradespeople started to find Marty and sign up. Some still do, every month, without me touching anything. That felt like progress.
It was not the progress that mattered. A marketplace needs both sides, tradespeople and clients, at the same time, in the same city. That is not an SEO problem. That is a sales, partnerships and cash problem. And every week I spent on it, I understood a bit better that I did not have the seller in me, and that I did not love this sector enough to grow one.
That was the hard part of the year. Not the work. Carrying a project that fit my former partner better than it fit me, with the pressure of a shrinking runway and nobody to share the decisions with. I was mentally exhausted by the end of the summer, and I am saying it plainly because it is part of the story.
The decision
By August the choice was clear. Either put 10,000 to 20,000 € into acquisition and give the marketplace a real launch, or stop.
I stopped, early September. Not because the product was bad. Because I could feel that continuing would have been about not admitting the year was over, not about believing in the business. Spending that money would have been ego.
The cost of the year, in round numbers: about 6,000 € lost on infrastructure, tools and running costs, and between 30,000 and 40,000 € of salary I did not earn. I am comfortable publishing those numbers. They are small compared to what I learned, and they are real, which is more than most founder stories offer.
Valyra, or fixing the part I was missing
I did not want the year to end on “I cannot sell”. So in early 2026, on evenings and weekends, I built Valyra: a pipeline that writes sector content where every claim is scored and traced back to an official source, INSEE, Légifrance, ANAH, and refuses to publish a sentence it cannot source. The idea was to turn the thing I had learned by accident on Marty, getting found through content, into something repeatable.
I tested it on Marty itself. The method worked: with very little time, the search traffic kept climbing, and the signups kept coming.
And then I was honest with myself a second time. The added value over a well-prompted Claude was thin. The people who needed it most could not pay for it. And I did not want to spend the next years in SEO, a field I respect but do not love. By then I was at Yubo, on a data platform I actually enjoy, and that settled it.
Both products still run. Neither is my job.
What I kept
Four things, in my own words.
Love the subject, not just the problem. I was the technical guy in a sector that was not mine. A good engineer can build for any industry, but building a company in one takes a kind of stubbornness you only have for things you care about.
Test the partnership before the product. We had done fifteen interviews with tradespeople and none with each other about what would happen if one of us wanted out. That conversation costs an afternoon.
You are solving a problem, not writing software. In a job or a freelance contract, being technically good is enough, because someone else has already decided what to build and why. On your own, nobody pays for a clean architecture. They pay for a problem that goes away. I knew this in theory. I know it in my hands now.
Tech is not the point. This is the same lesson from the other side. I love the craft, and the year did not change that. It changed the order of the questions. I now ask who a system is for, and what breaks for them if it is wrong, before I ask how to build it. At work, that is the difference between a pipeline that runs and a pipeline that someone trusts.
What it changed at work
I came back to a salaried job on purpose, and I chose the most demanding environment I could find. I am a better engineer than I was before the year, not because I learned more tools, but because I spent twelve months on the other side of the table, where the question is never “does it work” but “does it matter”.
If you are an engineer thinking about building your own thing: do it, with a time box and a number you are willing to lose. And when the product is finished, remember that you have just reached the start.