About
Siarhei Thor
I build applied AI systems for real use: LLM tools, retrieval workflows, automation, data pipelines, and ML-backed product features.
Why Serevion exists
Serevion is my public notebook for AI engineering: durable writing, lab notes, and small systems that make a technical idea prove itself outside a slide deck.
The recurring question is simple: what happens when an intelligent system meets real data, real users, and real constraints? Sometimes the answer is a useful product. Sometimes it is a beautifully confident mistake. Both are worth documenting.
About me
I am an AI/ML engineer in Zurich working on applied machine learning, LLM systems, and data-heavy product infrastructure. My current work includes end-to-end ML services for running and cycling analytics, wearable-data quality workflows, AI-assisted training feedback, and internal tools that turn scattered operational data into something a team can actually use.
Before AI engineering, I spent years in architecture and computational design, leading multidisciplinary building projects and using code to automate design workflows. That background still shows up in how I build software: make the structure visible, respect constraints, and do not confuse a polished model with a load-bearing system.
Sport is the other useful pressure test. I competed as an elite athlete in orienteering ( About Orienteering), and endurance sports are still part of daily life: road cycling, MTB, trail running, and whatever else creates a long enough conversation between body, terrain, weather, fatigue, and data. That is a good environment for learning what AI systems miss.
What I work on
- End-to-end ML services, from data preparation and model evaluation to APIs, deployment, monitoring, and retraining.
- Agentic LLM applications with retrieval, structured data, tool use, guardrails, and observable failure modes.
- Sports and human-performance systems built on wearable data, training load, and practical feedback loops.
- Personal lab projects such as Knowledge Hub, Crypto Lens, and research-radar workflows that test ideas in daily use.
- Plain technical communication that makes assumptions, trade-offs, and failure modes visible.
Find me elsewhere
The formal work history stays in my CV. Code and public projects live on GitHub. For professional contact, use LinkedIn.