15 projects across data, automation, apps, games and design.
Outbound campaigns needed clean, tiered company lists, but the raw PitchBook and Crunchbase exports were messy and overlapping. I wrote Python scripts that merge the pulls into one schema, clean scraped company names, deduplicate by LinkedIn company slug and then by normalized name, and build filtered lists per campaign. Each list is versioned with the script that rebuilds it, ready for Clay enrichment.
Farmers can't always afford to measure every soil metric. In this DataCamp project I trained one logistic regression model per feature (N, P, K and pH) and compared weighted F1 scores to find the single most predictive measurement for choosing a crop. It shows feature evaluation and multiclass classification with scikit-learn.
Does a Netflix title's release year relate to its user rating? I built a Streamlit dashboard that switches between a scatter plot with a regression trendline, average ratings by decade and by category, outliers and a correlation heatmap. It shows exploratory analysis and basic statistical testing presented in an interactive app.
Cancer outcomes vary across patient demographics and lifestyle. I analyzed records of 17,686 patients: cleaned them into ordered categories, visualized survival months by cancer type, gender and ethnicity, ran systematic sampling, and fit logistic regression models to predict recurrence. It shows independent EDA, statistical sampling and classification modeling.
Do women's international soccer matches have more goals than men's? In this DataCamp project I filtered FIFA World Cup results since 2002, found the goal distributions were skewed, and chose a right-tailed non-parametric test at a 10% significance level. It shows picking a statistical test based on the data's distribution.
Currency rate data has to be collected regularly, checked and stored before anyone can analyse it. I built a modular Python ETL pipeline that pulls the latest rates from the Open Exchange Rates API, validates and reshapes them with pandas, and saves them to CSV and SQLite, runnable once or on a schedule with a mock mode, rotating logs and pytest tests. It shows clean pipeline design, input validation and testable data engineering.