Built an LLM-driven EDA tool automating data cleaning, missing value detection, and visualization, reducing analysis time by 60%. Designed a multi-step pipeline to translate natural language queries into Python analyses across 10+ chart types. Automated bias checks and summaries, transforming exploratory workflows from hours to minutes.

Data Scientist at trivago N.V.

Reduced costs by ~10% and cut false positives by ~15% in accommodation-type grouping using improved differentiation strategies and LLM-based techniques. Improved duplicate/mismatch detection accuracy by ~12% with automated pre-filtering that reduced manual evaluation by ~35%. Evaluated ML and transformer models (XGBoost, CatBoost, TabPFN, Random Forest; BERTa, RoBERTa, DeBERTa) for detection tasks, achieving ~4% accuracy gains. Analyzed 500k+ user interactions from an A/B test on sponsored listings, contributing to an ~8% uplift in CTR and revenue potential. Analyzed 1M+ user sessions to compute KPIs (clickout ratio, engagement, conversion) and identified high-impact actions driving bookings, boosting conversion by ~5%.