20 years building systems. Lately, building systems that think. I'm Dominik Suwara — hands-on engineer, team leader, and a genuine obsessive about what AI does to us as humans, not just for us.
I've spent 20 years in the deep end of IT — data engineering, ETL pipelines, SQL Server, full-stack development, and leading teams that ship real products. I carry dual degrees in Computer Science, and I've been hands-on every step of the way.
As Director of Application Development at CLS Group, I work at the intersection of financial infrastructure and modern engineering. But the thing that genuinely excites me right now is what AI changes — not just for software, but for how humans think, reason, and make decisions.
I completed a Cambridge cognitive psychology certificate focused on natural vs. artificial brain reasoning — specifically on how we avoid AI-induced cognitive bias. That's not a side interest. That's a lens through which I look at every AI project I build.
British citizen, Polish roots. I build things on my own infrastructure, at my own pace, for real purposes. No toy projects. No demos for the sake of demos.
I think about AI the way a psychologist thinks about behaviour — not just what it does, but what it reveals about us. The questions that interest me most are at the boundary between artificial and human reasoning.
Everything I build runs on self-hosted VPS infrastructure. I control the stack end-to-end — compute, data, APIs. It keeps me honest about costs, latency, and what "production" actually means.
I don't theorise about what AI can do. I build it. Voice-controlled trading systems, autonomous WhatsApp agents, PDF-generating ebook platforms — all self-deployed, all real workloads.
The AI wave is real. But so are the risks — to autonomy, to epistemics, to what we outsource from our own minds. I hold both views simultaneously, and I think that tension is necessary.
Self-hosted, production-deployed, solving real problems. Updated as new projects ship.
AI-powered ebook generator sold on Etsy. Topics → structured chapters → formatted PDF with AI-generated cover art. FastAPI backend, WeasyPrint rendering, pgvector for semantic layers, Imagen 4 for covers. Full cost tracking per generation.
Natural language forex trade execution. Speak or type a trade instruction — it parses intent, validates parameters, generates a confirmation card, and executes on MT5 via IC Markets. Built with strict safety gates and 60-second auto-cancel logic.
Personal AI assistant via WhatsApp Business. Handles text and images, maintains 10-message conversation context, enforces strict 50-word response discipline. Grok (internet-enabled) as primary model, Gemini as fallback. Whitelist security.
Personal knowledge base system. FastAPI + SQLite FTS5 full-text search, git versioning for every document, AI-assisted writing via Grok, JWT httpOnly auth. My own self-hosted Notion replacement that I actually control.
md.dominiksuwara.cloudSelf-hosted AI agent platform running on Docker. WhatsApp interface, Google Drive / Gmail / Calendar integration, PDF and vision capabilities. Multi-model support with n8n for orchestration and scheduled tasks.
Production website for a Polish electrical contractor. Static site with AI chat bot restricted to company knowledge, admin panel, contact-to-n8n webhook automation, Google Maps, mobile-first at 360px base. Cloudflare Worker for secure API proxying.
rsinstalacjeelektryczne.cloudFastAPI + vanilla JS tool generating realistic synthetic forex trade data for ML training. 74-column trade schema, 6-state lifecycle simulation, Grok for natural variation. Writes directly to SQL Server 2022 — 5,000 EUR/USD trades generated.
This page updates whenever something ships. Check back.
Questions I keep returning to — not as abstract theory, but as things that affect how I design every system I build.
If a model confidently generates an answer you didn't know was wrong — what did your brain actually learn?
AI & Epistemic DependencyWe talk about AI alignment as if it's purely a technical problem. But misalignment between human cognitive shortcuts and AI fluency is already happening, right now, in every office.
Cognitive Bias in AI-Assisted WorkThe most dangerous AI isn't AGI. It's a mediocre LLM that sounds right 90% of the time — deployed without oversight, in decisions that matter.
AI Ethics in PracticeThere's something profound about training a machine on everything humans have written — and watching it produce something that feels like understanding, without the lived experience that gave those words meaning.
Artificial vs. Natural ReasoningThe same cognitive psychology that explains why we trust authority also explains why people trust confident-sounding AI. That's not a coincidence — it's a warning.
Psychology of AI TrustI want AI to augment what I think, not replace it. The moment I stop noticing what the model got wrong is the moment I've stopped being the engineer in the room.
Human-AI CollaborationI'm always interested in conversations about AI systems, data engineering, self-hosted infrastructure, and the human side of technology. Not looking to sell anything — just genuinely enjoy talking to people building real things.
I started with computers because I was fascinated by how they worked. I stayed in the field because I was fascinated by what they reveal about how we work.
AI is the most interesting chapter yet — not because of what it can do, but because of what it forces us to ask about intelligence, meaning, creativity, and what it is to understand something.
Everything on this site was built in that spirit. Nothing outsourced. Nothing theoretical. All of it running in production.