Data Scientist | Data Analyst | MSc Computer Science (Data Science & Technology)
I translate messy, real-world data into actionable insights through clear analysis, modeling, and practical prototypes. I care about human-centered technology and building decision-support tools that are rigorous, easily understandable, and above all, useful.
- Interests: product & business decision support, evaluation, applied ML, data visualization, LLM prototyping
- Open to: Data Scientist / ML / Analytics roles
Languages: Python, SQL, Java
Data: EDA, data cleaning, statistical analysis, KPI/metric definition
ML: regression & classification (linear + tree/ensemble), feature engineering, cross-validation, benchmarking/selection, evaluation, class imbalance handling
Libraries: pandas, NumPy, scikit-learn, XGBoost, GridSearchCV
Tools: Jupyter, Git, Docker (basic)
Visualization: Matplotlib, Streamlit, D3.js, Tableau
Applied AI: LLM prototyping (Mistral), prompt design & iteration
Collaboration: stakeholder communication, requirements gathering, clear reporting/documentation
Tech: Streamlit, Python, Mistral, cloud DB
- Built an LLM-powered chatbot guiding crowdworkers through SMART goal setting and twice-weekly reflection prompts
- Deployed with a cloud-hosted database and external GPU-based inference endpoint
- Implemented database-backed session context retrieval + logging for consistent UX and longitudinal evaluation
- Result: +26.6% increase in users’ goal attainment in the study
Links: Demo on Render | Paper on TU Delft repo
Tech: Python, XGBoost, SMOTE, Streamlit, Tableau
- End-to-end ML pipeline to predict churn and support retention decisions
- Benchmarked multiple models; selected XGBoost + SMOTE, optimized for recall using F2-score
- Performance: F2 = 0.71, Recall = 87%
- Presented insights via an interactive Streamlit app + Tableau dashboard
Tech: Python, XGBoost, GridSearchCV
- Predicted term-deposit subscriptions with a precision-focused objective to reduce wasted outreach
- Cleaned data, engineered features, benchmarked classifiers, tuned XGBoost via GridSearchCV
- Performance: Precision = 0.78, F1 = 0.65 (Recall = 0.56)
- Surfaced key drivers (e.g., month, job segment) + targeting/timing recommendations
Tech: D3.js
- Built an interactive dashboard to explore tourism trends across regions and seasons
- Designed filtering + comparison views for quick pattern discovery by non-technical users