Receipt Machine
Emailed receipts into the books. 30 s per document, no cloud.

AI · Agents · Automation
AI Automation Developer / Pardubice, Czechia
I build agents and automations that remove the whole task. Receipts read themselves in 30 s, leads find themselves, data stays in the company. n8n, local models in Ollama, my own RTX 3090. A person signs.
Emailed receipts into the books. 30 s per document, no cloud.
Agents make a video from topic to final cut. Budget near zero.
Pilot produced, October 2026
Searching crypto for an edge. V3: 8.8 million bars, a no-go written down first.
V3: 8.8M bars, 21 tests, a no-go written down first
My main client. I automate their administration over the long term, step by step and all on their own hardware.
A second, smaller collaboration that stays open: contract AI development for an engineering company.
90 minutes online for the cluster board: Copilot, Ollama, AI agents.
Trade licence since September 2026. Automation and AI training for companies, my own products on the side.
Client work since 2020: 3D, websites, photography. That is where the fixed deadlines and brief-to-handover habit come from.
Blender and client work since 2020, then n8n and local models from 2025. Self-employed in Pardubice since September 2026, alongside high school (IT at DELTA High School, Pardubice).
Today I build AI agents and automations for companies drowning in paperwork: documents a model reads, databases that connect departments, and agents that do the routine for people. Before I promise anything, I check that it works. And when it doesn't, I say so.
The machine drafts, a person signs. Sensitive data stays with the client and a local model reads it. I build small blocks instead of one big workflow, so each can be fixed on its own, and I write documentation the client can run it with, without me.
Automation for companies, AI training for teams, an internship, abroad too. Tell me what repeats at your place and where the data lives. I will answer and we set up a call.