AI Engineer — agentic systems, MCP, LLMs
ML Engineer · Data Scientist · Chemist (Diplom-Chemiker) · Maille Artisan
I build production AI systems: agentic frameworks, MCP servers, RAG pipelines, and the infrastructure around them. Creator of agent-almanac — an open-source library of executable AI skills, agents, and teams built on the Agent Skills open standard. CRAN package author (putior). 5 peer-reviewed publications spanning ML methodology and organic chemistry — including a cover-feature paper in Angewandte Chemie (2012, issue 30, DE + Int. Ed.) — 128 citations, h-index 5.
Daily driver: Claude Code with Opus — agentic MCP workflows across R, Python, Rust, and JS codebases. The open-source agent-almanac captures how I organize skills, agents, and multi-agent teams for this kind of work.
Available for 100% remote AI engineering roles within Europe. Relocating to Austria in Autumn 2026.
- Agentic AI frameworks — skill definitions, multi-agent coordination patterns, validation with step-level error recovery
- MCP server development — including cross-review-mcp for inter-agent artifact exchange
- Autoregressive → diffusion model conversion (
ez-ar2diff) — currently in stealth mode. Ping for a teaser. - GPU kernel engineering — hand-tuned CUDA/SASS kernels
- Production LLM integrations — RAG pipelines, embedding retrieval, Sentence Transformers
- Open-source R package development — putior (CRAN), jigsawR
AI daily driver: Claude Code (Opus) — MCP orchestration, polyglot agent-driven workflow Agentic / LLM: MCP, Anthropic API, Sentence Transformers, Agent Skills open standard, RAG ML: tidymodels, scikit-learn, xgboost, PyTorch, diffusion models Languages: R, Python, SQL, TypeScript, Rust, Bash Infrastructure: Docker, Linux/WSL, CUDA, nginx, FastAPI, DuckDB, Parquet
- jigsawR on Hugging Face Spaces — containerized demo of the R package for jigsaw puzzle generation
- Triple Pendulum — interactive Three.js/WebGL simulation of a chaotic triple pendulum, with accompanying dataset (55+ downloads)
- Hugging Face — spaces and datasets
- Google Scholar (128 citations, h-index 5)
- ORCID
- Kaggle
If my open-source projects save you time, consider sponsoring their development. Reliable AI assistance requires structured knowledge — and maintaining that structure is work the models themselves cannot do.