builder log_
A record of things I've founded or built and my learnings from them.
-
Engine
An old friend from Toucan convinced me to join him on Engine. We are building a fintech crypto platform where users can create their own ETFs or structured products.
I have learned a lot about financial engineering products, and I think they are amazing and beautiful. This is the first company I have done with more than two founders. There are four of us, each with a specialization. The output is much bigger, and the work and burden do not sit on one person. It is still ongoing, so there is no full conclusion yet.
So far, I think I want to do more of these. A few people who are very good at their thing can move much faster than one or two founders trying to do everything.
-
WalkUp
Me and my wife are big on walking, we barely touch our car most weeks. In early 2026 we've decided that we want to move to a different city for various reasons. Not knowing the city well, we were in a bit of a bind: what locations are walkable there?
To make the process easier I've built WalkUp. It ingests and evaluates all sorts of data, like points of interest, slope, trees and so on to produce a walkability score that is actually accurate (unlike Walk Score). Fun fact, I had to time my walks to various places in my hometown because most maps apps get walking distance wrong (due to semaphores mostly).
During the process I've also interviewed some real estate agents and I've learned about the concept of MLS software. This could be an idea of a B2B offering to them, but I didn't feel like there was market-founder fit here. For me the win was that we managed to find a place to rent where we loved living there.
Dogfooding is imperative if you want your product to be genuinely useful.
-
Job Finder
After re-entering the job market I got tired of clicking through careers pages. I went on and built an agent that scrapes ATS platforms and evaluates each job post with a multi-stage keyword and LLM pass against my target profiles, and when it finds a job that is likely to be a fit, it puts it in a review queue for me to take a look at and apply to.
The initial version had a pretty high false positive rate (30%), but over time I've managed to build an eval dataset of over 100 real examples of jobs I liked and didn't like, with the reasoning for that, and I have managed to get it to a false positive rate of about 10%.
And before you ask, yes this is actually working. I have had multiple interviews from this thing.
You absolutely need evals if you're going to build an agent, and it is absolutely tedious to build your eval dataset. I found it easier to build it from usage instead of trying to build it up front, so now all of my agentic applications have forms to provide feedback on the various AI-generated content. I guess this goes back to dogfooding?
-
Zap
A LeetSoftware client launched XGame, an EVM gaming dApp, and went broke because he kept all his money in the game token instead of stablecoins. After launch, KOLs dumped the token, the price slid, the community got angry, and players left within two weeks. It was a mess. The community was insulting him and he had to cash out and close the shop down.
The code and the product were genuinely good and had a real edge, so a number of us took the project over. We rebranded it to Zap and came up with some novel ideas for new B2C games on top of the existing crash game. We came up with the novel idea of offering white-label iGaming infrastructure coupled with crowd-funded bankroll. It had real potential.
I brought an angel from my network who was ready to put in $100k. On our first call he was prepared to send the money on the spot if we had a SAFE lined up. I was a minority co-founder and the major co-founders moved too slowly, so he got cold feet and never invested. We kept trying other sources, nothing landed, and even though the idea was good, obviously profitable, and we even had a gambling license, the whole thing died a slow death.
When something is hot, you have to close fast. Have the SAFE and the bank account ready on the day you pitch. Fundraising, and a lot of business deals, run on gut feel in the moment, so strike while the iron is hot.
-
Tech recruiting
After helping build half the dev shop team, I thought the right business might be recruitment or staffing. I could meet solid developers, filter them well because I am a developer, and had already placed five people within a few months for our dev shop.
It was very hard to find employers who wanted to work with a recruiter and pay a fair finder fee. The companies that did often wanted brand names or a long interview process. I could tell quickly whether someone was good enough for a startup with 5-20 engineers. Many of my clients could not.
A profitable recruitment shop is a marketplace. You need a big bag of candidates and a big bag of employers, then a small percentage match. I have not fully given up on this one. I may come back in an AI-augmented way.
-
Dev shop
I tried to build my own dev shop. I made a website, spoke to four or five developers about freelancing as subcontractors, and started reaching out to sell the work.
It did not work out. Finding clients was much harder than I understood then, especially at prices where I could make money and still pay developers enough to keep them happy. Turns out to have a healthy agency business you have to markup your workforce by ~70%-100%. I later joined a dev shop as a partner and spent a year helping build its team. That shop is now LeetSoftware, so this story got a second life.
The hard part (for me) is not finding developers. The hard part is finding clients at a price that makes the business work.
-
Minerva
A friend and I saw the RAG and connector wave coming. Companies wanted to connect Google Drive, Notion, and other internal information to an LLM, but many did not want to give that data to a connector vendor. We made Minerva, an open-source, self-hosted API that turns S3 documents into a searchable knowledge base.
We built the MVP and had a few calls. There was some interest, but we could not sell it for real. In hindsight, the business would have been killed anyway. Claude now has connectors and MCPs.
It was an interesting experiment and my first real foray into AI, and into more aggressive agentic coding.
-
Expensio
Expensio started as a quieter way for me to track my own expenses, income, and budget. I made it self-hosted because I had a bet that self-hosting and one-time purchases would become a big thing.
This was where I experimented with SEO and free tools, which taught me useful things. The product itself did not get traction, so I shut it down.
People do not want to self-host. They want a done-for solution. The people who are going to pay want to give you money and have the problem go away. Also free public tools with input params in the URL are great for SEO.
-
Nymnio
Nymnio was accounting software for crypto companies. It tracked transactions across addresses, cash flow, income, expenses, and capital gains. With some labelling help from the user, it could export the statements that made bookkeeping much easier.
I spoke to a few accounting shops. They liked the idea and were probably willing to spend a few thousand dollars a month on it. It did not die because the idea was bad. I just did not have the time to take it to the end, so it slowly died off.
When you see a B2B opening and a few people are actually interested, that is a very good sign. Move much faster. Give them what they want ASAP!
-
Pickle Farm
While I was at Toucan Protocol, I built an MVP for a casual idle play-to-earn game. You would come in a few times a day, plan things, set things to pickle, then sell the mature pickles on the open market for pickle coin.
I got the architecture working, made visual assets in Aseprite, and talked to some people. By the time I had something to show, though, the NFT craze had already passed. I saw that it was not going to get major traction and stopped.
If you are going to ride a trend, you have to ride it hard and early. If I could go back, I would launch an NFT with Pickles first, then build the game on top of the created community.
-
Blockchain Creators
I built a Gumroad or e-junkie-like product that let course creators take payment directly in crypto. I had spent a lot of time as a freelancer building websites and membership sites for creators, and many of them had started investing in crypto. One of them, a client making about $1M a year selling courses, genuinely wanted to sell his courses for crypto, so he invested $30k for 40% of the company. I called it Blockchain Creators, and I even had the blockchaincreators.com domain.
I was a year and a half into my career and knew nothing about startup valuations, so I priced the deal by my own runway: $2,500 a month to build full-time instead of freelancing. He paid in monthly installments so he had some assurance I would keep building. I did not know installments were not standard either.
The product technically worked but the business model didn't. He became my first user and beta tester, which made the product better, but I've learned that even from a creator actively motivated to sell for crypto, almost none of his buyers wanted to pay in it. After months of DMs and content, I had about a dozen users and a few hundred dollars a month. I stopped the agreement and walked away from the remaining money. It was not a fun decision but I felt it was the right decision.
It is not actually that crazy to get an angel investor. Getting good terms is another matter. Even the most motivated first user could not make his buyers want crypto, so test the hypothesis before you scale. The idea was a Gumroad feature, not a company.
There's probably at least a few others I'm forgetting tbh.