All Projects

← Back to Portfolio
OLIST
SQLPYTHONA/B TESTINGPOWER BI
OLIST Marketplace Analysis
  • A 97% one-time buyer retention crisis was hiding inside 99K orders on the Brazilian OLIST marketplace. I built a full medallion data warehouse (Bronze/Silver/Gold) in MySQL and segmented 93K customers into five RFM behavioral groups.
  • Then I designed a statistically rigorous A/B test validating a coupon intervention at p<0.001. Outcome: Identified R$4.6M retention opportunity and validated a retention strategy achieving 186% ROI.
View Project
FMCG
EXCELPOWER QUERYVBAPOWER BI
FMCG Sales Automation
  • A regional FMCG distributor was spending 80-115 minutes daily on manual sales consolidation and reporting—a workflow bottleneck that delayed business decisions and created error risk.
  • I built a fully automated pipeline where Python generates daily CSVs at 6:30AM via Windows Task Scheduler, Power Query ingests and cleans all files automatically, and Power BI serves a single-page executive dashboard. Result: 30-40 hours/month of manual work eliminated with zero daily intervention needed.
View Project
AWS healthcare architecture diagram
AWS S3GLUEATHENAQUICKSIGHT
AWS Healthcare ETL Pipeline
  • Raw hospital encounter data (~100K records) was inconsistent and unqueryable for operational decisions. I built a layered cloud ETL pipeline on AWS with S3 raw/cleaned/curated zones and Glue DataBrew for profiling and cleaning.
  • Glue PySpark jobs handled transformations and aggregations, Athena supported SQL validation, and QuickSight delivered stakeholder dashboards. The pipeline surfaced actionable KPIs on patient volume by department, average length of stay, and admission type distribution.
View Project
Japanese money for loan default risk analysis
EDARISK ANALYTICSPYTHON
Loan Default Risk Analysis
  • Conducted exploratory data analysis to identify signals associated with loan repayment difficulty for a consumer finance company.
  • Analyzed applicant demographics, credit history, employment patterns, and loan characteristics to surface risk indicators.
  • Deliverables included risk segmentation framework and data-driven lending recommendations to support risk-aware credit decisions while avoiding unjust rejection of capable borrowers.
View Project
Image by Egor Komarov for web scraping project
PYTHONBEAUTIFULSOUPAPI INTEGRATION
Web Scraping & API Collection
  • Built three reusable data extraction workflows to gather structured data from web sources for analysis: (1) E-commerce product scraping using requests + BeautifulSoup, (2) Social media data via RapidAPI integration, and (3) Real-time stock market data collection.
  • These pipelines demonstrated data acquisition techniques essential for analytics projects where clean datasets aren't readily available.
View Project
Image by Hartono Creative Studio for fake news detection
NLPWORD2VECMACHINE LEARNING
Fake News Detection
  • Built a binary classification system to detect fake vs. true news articles using Word2Vec embeddings and supervised learning.
  • The model analyzes textual patterns, writing style, and linguistic features to identify misinformation with 92% accuracy.
  • Demonstrates NLP fundamentals and understanding of how text data can be transformed into numerical representations for machine learning.
View Project