Threadline
Fashion e-commerce analytics for 31,047 order lines of a Pakistani clothing brand. Threadline audits every column before trusting it, finds that 4 of the 18 columns were made up and that the "seasonal trend" was a date bug, then shows where the store really loses money in a live dashboard and a 10-page Power BI report.
- Python ETL with 17 data quality checks and a verdict for every column, loaded into a PostgreSQL star schema
- Data forensics: channel, age, gender and customer ID were randomised, and day and month were swapped in the date column
- 16 business questions in SQL with CTEs, window functions and a market basket self-join
- Finds that 322 of 1,164 styles make 80% of sales and that items under Rs 1,000 fail 16.5% of the time against about 6% above
- A LightGBM returns model (AUC 0.737): the riskiest 10% of lines fail 3.4 times more often than average
- Twelve-page React dashboard: cross-filtering explorer, plain-English Ask box, order inspector, Pakistan map, SQL Lab and Excel/PDF reports
- A 10-page Power BI report generated from code, with 65 DAX measures, a metric switcher, what-if slider, drill-through and report-page tooltips
Built with
- Data & ML
- Python, pandas, LightGBM, scikit-learn
- Database
- PostgreSQL, DuckDB, Advanced SQL
- Frontend
- React, Vite, Tailwind CSS, Node.js
- BI
- Power BI, DAX