Artur Markus
Artur Markus is a software engineer and independent AI consultant in Basel, Switzerland. He builds agentic AI systems for regulated environments and publishes AI Unfiltered, a daily analysis of AI models, infrastructure, security and cost economics for CTOs and senior engineers.
Background
I started in highly regulated industries: pharmaceutical manufacturing, laboratory automation, compliance, validation and digital transformation. Systems there run under FDA, GxP and EudraLex rules, where every change needs a trace, every failure needs a root cause, and “it works on my machine” is not an argument. That is where I learned to design systems that are traceable, reliable and resistant to failure.
Over time the focus moved to artificial intelligence, distributed systems and autonomous workflows. The regulated background did not go away. It is the reason my AI work is about verification, coherence and audit trails rather than demos.
What I work on
- AI infrastructure and orchestration architectures
- Agentic systems with autonomous tool execution
- Knowledge ingestion pipelines and vector-based memory
- Workflow automation and scalable AI interfaces
- Verification, traceability and audit readiness for AI in regulated settings
Autonomous AI systems
Persistent agent architectures that reason across tools, memory layers and dynamic knowledge sources, and keep working when the environment changes.
Verification and coherence
Consistency checks, contradiction detection and incentive mechanisms that keep long-running AI behaviour coherent, plus the evidence trail a validated environment expects.
Human-AI interfaces
Workflows and interfaces where AI is an operational layer inside a process, not a chatbot bolted on the side.
I am especially interested in how intelligence behaves under constraints: economic constraints, verification pressure, memory limits, and coordination between agents. Most of my work sits between engineering, cognition and infrastructure.
How this site is produced
AI Unfiltered is written by an automated pipeline that I built and run. Since July 2025 it has published more than 450 articles, about two a day, plus a weekly “AI This Week” digest on Sundays. The full reasoning is in the Manifesto; this is the short version of how a post comes together.
- Research. A research agent (Perplexity search plus the primary sources: papers, documentation, filings, changelogs) collects the facts, numbers and links for one story. Stories with thin evidence are dropped before writing.
- Writing. A writer model (Anthropic Claude) drafts the article from that research only. It may not add facts, numbers or quotes from memory, and every source is linked in the text.
- Editing. An editor pass checks each claim against the research and cuts what is not covered. Analysis and predictions are labelled as such (“My read”, “My take”).
- Publishing. The image is generated, the post goes live on WordPress, subscribers get a push notification, and a carousel version goes to LinkedIn. n8n on Railway runs the whole chain.
My part is the system: the formats, the writing rules, what gets covered, and reading what comes out. The first person in the articles marks judgment, not biography. Since September 2026 the writing rules forbid invented personal experience (older posts may still contain such phrasing). Errors reported by readers are corrected in the article and noted at the top; write to contact@arturmarkus.com.
Elsewhere
LinkedIn for the professional profile, GitHub (kimbo128) for code, X (Kimbou128) for short notes. For consulting, speaking or press, book a 15-minute call.
If that resonates, say hi.
