Somewhere between "this shouldn't work" and "wait, why does nobody else do this" — this is where I build.
I design AI systems for e-commerce and SaaS brands who are very happy with the results and very nervous about you finding out.
28 brains.
40+ workflows (not these 280 free ones — mine have been yelled at, rewritten 6 times, and run 24/7 on real money. Reluctantly.)
Here are some cases with the names scratched off.
Pick the one that matches your current mood.
The awkward welcome speech
Who I am, why I built this, what you're looking at.
How it actually works
Architecture walkthrough — agents, scoring, reject loops.
What happens when nobody's watching.
Case study where I can't scratch the name off because the client is me, the talent is AI, and the CEO is a rabbit.
Before we go further, a plumber.
Two encyclopedia entries for the same profession
Plumber, Theoretical
A tradesperson who, when asked to fix your sink, explains siphon physics, P-trap geometry, vent stack dynamics, and the fluid mechanics of negative pressure — while you nod politely, understanding nothing.
Plumber, Practical
A tradesperson who, when asked to fix your sink, looks at it, fixes it, and says "nothing leaks, enjoy your bath 24/7."
§ AI Vendor, Contemporary — almost always of the theoretical variety. Talks RAG pipelines and multi-modal orchestration to hide the fact they built a ChatGPT wrapper. The following entries translate them back into human.
Everyone says agentic.
Here's what it actually means.
Five common terms, translated
The difference isn't AI. The difference is architecture.
Compound systems with scoring gates and reject loops don't just "use AI" — they build an intelligence layer that gets measurably sharper every week. That's not a feature. That's a moat.
How to build systems
that actually think.
TRIZ
Inventive problem-solving. 40 principles for resolving contradictions without compromise.
Cynefin
Sense-making framework. Classifies problems before choosing solution strategies.
OODA Loops
Observe-Orient-Decide-Act. Speed of decision as competitive advantage.
Dual-Process
System 1 vs System 2. When to let agents think fast vs think slow.
MariaTukhas
10 years of marketing — agency, then products, then AI systems that run them autonomously. The path made zero sense until it did.
Stopped hiring humans. Started architecting machines. The machines haven't complained yet. The ones that did got deprecated.
"I build systems that learn, not prompts that guess."
I like interesting projects and unique challenges.
Also adventure.
Adventure I like a lot.
If you've got one — you know where to find me.