I spent my early career in banking, where I made my first steps in reporting. After that I moved into the outsourcing sector, where I extended those reporting skills and operated in a multinational environment. The last few years I’ve been pivoting into Machine Learning and AI automations with agentic workflows.
Stack I work with day to day: Python, PyTorch, TensorFlow, scikit-learn, LangChain, LangGraph, LlamaIndex, MLflow, AWS, Docker, Terraform, GCP, PostgreSQL, MySQL, MongoDB, QuantLib, Jenkins, Streamlit, Tableau, Power BI, Excel, n8n, Airtable, Notion, Google Analytics, Claude Code, Jira.
Production screener over ~2,000 US equities. Daily pipeline into MySQL, with fundamentals, technicals (RSI, moving averages, analyst targets) and interactive Plotly charts. Agentic workflow pushes signals from database to Slack. Running live on its own domain.
View project →Quantitative deep-dive on 12 equities: Black–Scholes option pricing, Sharpe ratio, probability of loss, and derivative-based summary statistics. The math foundation behind the ML projects above.
View project →Supervised regression on the German real-estate market. Feature engineering, model tuning, and an iteration log showing how the model improved version by version — not just the final number.
View project →Transaction-level retail modelling with gradient-boosted trees. Three model iterations, each documented with what changed and why — the same iterative loop I follow on every project.
View project →End-to-end deep learning system: ingest equity price history, train sequence models (LSTM and Transformer), serve predictions through a FastAPI backend, and visualise them in a live dashboard.
View project →Upload a book as a PDF and get chapter-by-chapter summaries — detailed, brief, or key points — plus chapter-scoped Q&A. Built on chapter chunking instead of blind semantic search: PyMuPDF pulls chapters straight from the PDF (no manual Markdown step), embeddings land in pgvector, and summaries are cached in Postgres so each one is generated once.
View project →