• Architected an Agentic information retrieval system to automate legal research across 175,000 Swiss statutes and 2.47 million federal court decisions. • Tuned Qwen2.5 model using Unsloth and evaluated it with vLLM, achieving a 94.4% MAP@3 score. • Used xmltodict and PyPDF2 for data extraction from XML/PDF files to analyze data references in articles. • Engineered and fine-tuned a XGBoost model to predict customer churn, achieving a 97% F1-Score. • Built a LightGBM model to forecast monthly sales with a 8% MAPE, improving inventory management. • Stack: Python, TabPFN, Tabular, Qwen, Unsloth, vLLM, XGBoost, LightGBM, NLP, RAG

• Delivered ML forecasting models are leading to an estimated C10M yearly saving by improving inventory management. • Built internal package with 15+ reusable components, reducing code duplication in forecasting workflows by 30%. • Created 20+ Pytest modules validating 6 pipelines, preventing 95% of data quality issues within ETL pipelines. • Scaled automated forecast validation and real-time Tableau dashboards to serve 4,000+ global clients. • Stack: Python, SQL, PySpark, Pytest, Tableau, Nixtla, Databricks, Forecasting, Time Series

• Analyzed e-commerce datasets to identify behavior patterns, driving a 10% increase in sales revenue. • Designed robust sklearn pipelines utilizing XGBoost and LightGBM for time-series forecasting. • Delivered 10+ end-to-end data science solutions for global clients, maintaining a 100% success rate. • Stack: Python, Scikit-Learn, XGBoost, LightGBM, SQL, Analytics