Hi, I'm Thomas.

I combine a biomedical research background and more than five years of clinical data management in a SAS- and SQL-based research environment with recent hands-on training in AI Engineering and Data Science.

My focus is on applied AI and machine learning for healthcare and other data-intensive or regulated environments. I am particularly interested in medical NLP, retrieval-based systems, structured clinical data, model evaluation, and deployment-oriented ML workflows.

• Dr. sc. ETH in Biochemistry
• Professional background in clinical data management and biomedical research
• Completed the neue fische AI Engineering & Data Science bootcamp
• Built two capstone projects involving biomedical information retrieval, medical NLP, RAG, model evaluation, APIs, databases, containerization, and monitoring
• Currently looking for junior roles in AI Engineering, Applied Machine Learning, Data Science, or Data Engineering

Experience with structured and clinical data, data validation, exploratory analysis, feature engineering, classical machine learning, model evaluation, reporting, and visualization.

Hands-on project work with biomedical text, embeddings, semantic and lexical retrieval, hybrid search, RAG, cross-encoder reranking, local LLMs, and reproducible evaluation.

Experience building project-based end-to-end ML workflows with APIs, relational and vector databases, testing, experiment tracking, containerized services, monitoring, and deployment.

Daily work with Git and GitHub, VS Code, command-line tools, notebooks, scripts, technical documentation, and AI-assisted development.

A completed AI Engineering capstone exploring the assignment of free-text medical-device safety reports to standardized MedDRA Preferred Terms.

The system combines:

• lexical and vector retrieval in PostgreSQL
• reciprocal-rank fusion
• cross-encoder reranking
• a local Qwen LLM through Ollama
• a FastAPI service and reviewer workflow
• MLflow-based experiment tracking
• Grafana monitoring
• pytest and GitHub Actions
• human-in-the-loop review and auditable result storage

The project was developed as a portfolio system for regulated-data scenarios. The original Hetzner deployment is deliberately offline, while the source code, architecture, evaluation results, and presentation remain available.

A team capstone focused on biomedical literature retrieval and question answering using PubMed Central content.

The project included:

• biomedical document processing
• keyword prediction
• BM25 lexical retrieval
• FAISS-based semantic retrieval
• reciprocal-rank fusion
• retrieval evaluation
• LLM-supported biomedical question answering

n8n prototype that checks vendor security advisories against an asset inventory: an LLM extracts the facts, fixed rules decide. Webhook intake, PostgreSQL, structured bypass without LLM, verbatim evidence check, 15 synthetic test cases, Docker setup.

Prototype of a tool that describes dev workflows as a DAG with automatic acceptance checks per step and runs them with Claude Code. Engine and CLI are covered by 161 tests; the sandbox run (Docker + git worktree per node, merge back) was tested on one server.

This repository documents a selected bootcamp exercise. It is separate from the two main capstone projects.