Software Architect · B2B · Poznan
I usually arrive where something already exists but stopped keeping up — an abandoned backend, a no-code app blocking growth, a process done by hand for years. I stabilize, modernize, automate.
No spec and no technical background needed — describe what stopped working, what takes too long, or where the business loses time.
The previous developer vanished with the source code. The business kept running 18,000+ orders across three sales points on that system.
An 18,080 CHF payout across 212 records silently completed only halfway — in a system running on real money 24/7.
A booking system for dental clinics: multiple locations, a live calendar and online payments — built from scratch.
Client feedback
The strongest proof is in the case studies below, but the client view belongs near the top of the page.
Studio Manager
“We started with one Telegram failure. Artur quickly understood the system and helped us organize payouts, employee access, notifications and automations. The system is more stable today and easier for the team.”
Owner of a service studio, Switzerland
Case study
Modernization and migration
“The program works perfectly, and we are also saving money thanks to the server change.”
Jerzy Czupik, owner of a multi-location retail company
Case study
Project proof
I do not lead with technology for its own sake. First comes the measurable result, then the technical move that made it happen.
~15 s → <1 s
Order printout time after replacing an abandoned Java backend with ~150 lines of Node.js.
Technical move
API reverse-engineered from HTTP traffic and the database schema, new Node/Express backend, React frontend untouched.
126 s → 3.9 s
An 18,080 CHF / 212-record payout that used to silently complete only halfway.
Technical move
Bulk operations (~272 HTTP calls → ~10), payout state machine, nightly audit + Telegram alerts.
6 deploys / 8 weeks
From a 450 PLN webhook fix to being a Swiss studio’s de facto technical partner.
Technical move
3 Telegram bots, 2 production systems, RBAC, a Deno Deploy microservice bypassing no-code platform limits.
Migration
14 minutes
Production cutover to new infrastructure — in a weekend window, zero data loss.
Hetzner Cloud, Caddy + Let’s Encrypt, daily off-site backup in another country with verified restore.
Case study ->Booking SaaS
9/9 E2E tests
A SaaS booking core with no double-booking, verified on production.
Multi-tenant RLS, atomic lock via a Postgres partial index, Stripe with an idempotent webhook.
Case study ->Scientific software
~460k → API
A probability grid of dynamical interactions from star-cluster simulation data.
Python pipeline, 16 grids at 20x20x20x20, 10k-sample round-trip validation, code being integrated into binary_c.
Case study ->AI lead gen
100 listings / $0.30
An AI pipeline that finds companies hiring people for work a script should do.
Rule-based pre-filter, Claude Haiku scoring, Sonnet personalization — full run in 2 minutes.
Case study ->How I enter a project
Four scopes, one working style: cut risk quickly, show a real result, then expand only where it makes sense.
Critical financial bugs, access control gaps, failing integrations — including where the no-code platform says "impossible". I isolate the root cause, not patch symptoms.
Outcome
Stable production without rewriting everything from scratch.
Previous developer vanished with the code? The database and network traffic are enough to rebuild the backend — cheaper and faster than a full rewrite.
Outcome
Recreate behavior first, then decide what deserves a rewrite.
LLM pipelines on real client data, not demos. Classification, extraction, transcription — the processes that eat a salary every month.
Outcome
LLMs work with data and constraints, not only prompts.
From spec to production in weeks, not months. Often as one app that works on every phone and computer — instead of three separate ones, at a fraction of the price.
Outcome
A small scope that can still be clicked and verified on production.
Systems from scratch
Not every project starts with rescuing production. Sometimes the right move is a small, concrete system: one process, one data model, quick proof the idea is worth pursuing.
3 work samples
2 public demos
1 scientific repo
Monteurzimmer match
A construction ZIP code -> nearby rooms, filtered by price and distance, with a German inquiry template ready to send.
14:30 / clinic A
Multi-tenant booking with database-level data isolation and a slot hold before Stripe payment.
20^4 probability bins
A bachelor thesis pipeline turns raw CMC simulation interactions into probability grids and an API for further calculations.
About the builder
I like the work that requires taking an existing system apart, understanding how it really operates, and putting it back together so it can keep evolving. Day to day: Node.js, Python, databases, integrations; on the AI side — agents and pipelines on real data, not demos. Mathematician from UAM.
Software architect
I stay after deployment
A system only starts living when people use it under pressure. SMS answered within the hour, fix shipped the same day.
Honest quote in 48h
The real number upfront instead of a cheap offer that ends in "phase 2". Full transparency — repo access, progress reports.
Full accountability
You talk to the person who writes the code and makes architectural decisions. Zero middlemen, zero lost context.
The audit is a starting point
Not a prelude to a big project. Sometimes the most honest answer is "this is not worth automating" — and that is the answer you will get.