AI Engineer (R&D)
I've been running this blog for a while and was never quite sure how to describe it. Not because I don't know what I write about — more because the topics I care about don't fit in one box.
Professionally, I work on AI systems. I try to understand what they actually do. I like looking under the hood — checking what's beneath the layer of abstraction someone carefully prepared for you. I spent a lot of time in cybersecurity, and you learn quickly there that what a system shows on the outside and what it does on the inside are often two different things.
I'm a practitioner, but I don't run from theory. Quite the opposite — practice without theory is just a collection of solutions without understanding why they work. That's why I write just as naturally about pipelines and architectures as I do about things that seem to lie somewhere else entirely.
When I look at economics, I'm interested in mechanisms, not opinions. How resource allocation actually works, where and why the system gets "stuck", whether it can be designed better. I'm a Marxist, but I treat it primarily as an analytical tool — a framework that lets you see things you wouldn't notice without it. I believe a planned economy is feasible at today's technological level. For me, this isn't a political thesis, it's an engineering one.
I also write about less systemic things. About projects I make. About things that irritate me. About observations that don't fit anywhere else.
I don't know if this is a blog about technology, or about society, or something in between. Probably about what happens when someone can't stop asking how things really work.
The entire Hermes Agent project is unresponsive, unintuitive, unsafe, vibe-coded — do not entrust it with any confidential information, I advise against buying a subscription or hosting it locally. It is a really poor project.
Eliminate the overhead and lack of idempotency of a full Windows system in automated testing. A practical guide to configuring WinPE and QEMU for lightning-fast boot, automating kernel debugging, and avoiding KDNET pitfalls.
I write about the recent updates to this blog and the decision to shut down the 'Agitka' editorial project, reflecting on the challenges and lessons learned from that experience.
I present a practical demonstration of how the integrity of a language model can be compromised during inference by modifying quantized weights in a GGUF file without restarting the server or injecting code. This highlights the importance of considering model storage and memory mapping in LLM security.
I write about how I discovered a vulnerability in the Wirtualna Polska (WP) mail service, which allowed me to execute stored XSS in user session. I explain the technical details of the bug, how I exploited it, and the response from WP's security team.