About
Sometimes existing software works perfectly.
Those times are boring.
That's why TachiBot exists — a platform that runs multiple AI models in parallel because one AI making mistakes isn't enough; I need six of them arguing about the best solution.
The Why
Named after Ghost in the Shell's Tachikomas because distributed problem-solving beats single points of failure.
What I build
I'm a product engineer who ships AI tools. Most of my work has been making AI less annoying to use — RAG pipelines, voice systems, multi-model setups. The models are impressive. Making them feel good to use is where it gets interesting.
TachiBot is what I'm building now — an MCP server that coordinates multiple AI models so they can argue with each other instead of just with you.
React, TypeScript, Python — whatever gets the thing out the door.
Outside of work, I play games that reward teamwork and strategy — basically the same thing but with better loot.
My French bulldog Nori watches me work with the quiet confidence of someone who knows she'd be better at it — if only she had thumbs.
Nori, Chief Supervisor of Operations
Unreasonably particular about
Darjeeling tea(first flush only)
D&D peaked with 5th edition(fight me)
Manhwa over manga(full color hits different)
Coding philosophy
Mostly use the tools that already exist, unless it's 3am and I'm convinced I can build a better version.
Spoiler: sometimes I actually can.
How It Started
"Surely coordinating AI models can't be that hard."
LLMs are librarians with an infinite library. Having all the books isn't the problem—knowing which ones to pull is.
Most systems just sort the haystack. TachiBot calculates the coordinates of the needle: craft the right prompts, gather the data, sort through competing options, search for patterns, reason through tradeoffs, and stress-test every answer to find the holes.
Data first. Then think. Then judge.
Location & life
Based in Gdańsk with my wife and Nori, where the pierogi are excellent and the Baltic is freezing.
Always up for discussing why your tabletop game needs more dice, why manhwa color palettes are superior, or why my AI models have stronger opinions about tabs vs spaces than I do.
Check out what I'm building
P.S. My best thinking happens mid-leash. Her verdict on the output: "Boring. Walk faster."