I spent around 10 months building TheoryCraft before I finally shipped it.
That sounds clean when you say it like that, but the real version is less pretty: I built too much, waited too long to market it, then needed an 18-day Ship or Die deadline to force the MVP out.
Still… it shipped. First real app out there. $0 MRR, 2 users, roughly 30 pre-registrations, and now the part I ignored for too long starts for real.
WHAT SHIPPED
TheoryCraft is the product I wanted for one simple question: would this trading idea actually work?
Not “does it look good on a chart”. Not “can I convince myself it should work”. I wanted a place where I can describe an idea, run it in a notebook, look at the code, inspect the results, and decide if the edge survives real data.
The stack ended up heavier than a tiny MVP:
- Phoenix LiveView for the app
- Postgres for product data
- Python and Jupyter for research notebooks
- gRPC for runtime boundaries
- Docker and Fly.io for exploration sessions
- MCP so external agents can work inside the product
The MCP part is still the thing I care about the most. I don’t want AI to give me a cute paragraph about a strategy. I want it to create a notebook, run the research, edit the code, show me the result, and leave me with something I can inspect.
That’s the product.
THE NUMBERS
Current state:
- $0 MRR
- 2 users
- around 30 pre-registrations from the launch window
- first launch sent by email
- Product Hunt still not the real growth engine
- SEO/GEO basically starting from scratch
The SEO numbers are ugly too. DR 2.3, 20 clicks in 3 months, 0.3% CTR, and depending on the tool, either 0 AI citations or 34.
I don’t really know which one to believe yet. But I know one thing: if people and AI systems cannot find the product, the build does not matter.
THE HARD PART
The technical part was long, but it was not the scary part.
I know how to build. I’ve been doing that for 10 years. The uncomfortable part is that I spent most of those years thinking the product was the only thing that mattered.
For TheoryCraft, that mistake is pretty obvious now. I had the platform, the notebooks, the MCP layer, the AI research flow, and almost nobody knew it existed.
That is not a product problem anymore. That is distribution.
And honestly, switching from builder brain to marketer brain is way harder than I expected.
WHAT HAPPENS NEXT
The next phase is not “add 20 features”.
The next phase is making the product discoverable:
- a real blog
- public research articles
- clean sitemap and robots.txt
- llms.txt
- comparison pages later
- Google and Bing indexing
- AI citations
- actual backlinks instead of pretending they appear by magic
This post is part of that. Not because a blog post will magically fix everything, but because I need the product to stop living only in my codebase.
The project page is here if you want the technical snapshot: TheoryCraft
Now I have to do the part I avoided: make people find it, understand it, and eventually pay for it
