I make messy spreadsheets tell the truth.

What I built

• Sales dashboard: sales, profit and quantity with Jan–Dec trend lines and % difference vs the previous year, sales and profit by segment, and weekly trends against an average line.
• Customer dashboard: total customers, sales and quantity per customer, customers by number of orders, and the top 10 customers by profit.
• Teal always means good and orange always means bad, so the dashboards read without a legend.

A problem I solved

The gold layer had no Order ID, so I joined it in from the silver layer on Customer ID. That paired every 2016 sale with all of that customer's orders from every year and inflated the order counts. I took the order date from the same table as the Order ID and counted distinct orders, and the distribution now adds up to the 325 customers of the year.

The challenge

Turn a single flat retail file (Central Superstore) into a proper analytical warehouse: a documented, repeatable load process and a model that business questions can be asked against.

What I built

• Excel → CSV conversion script in Python, then BULK INSERT into a raw Bronze layer, with truncate-and-reload stored procedures for Bronze and Silver.
• Gold layer as views: dim_customers, dim_products, dim_order_details and fact_sales, with surrogate keys and documented naming conventions.
• Data quality checks for each layer, a gold-layer data catalog, and a star schema diagram.
• Exploration and analysis scripts (change over time, cumulative, performance, part-to-whole) plus report_customers and report_products views covering recency, average order value, segmentation, and lifespan.

What I learned

The source data turned out to be already clean, so Silver stayed close to Bronze and only added a derived total_price column. Checking first and keeping each layer's job honest mattered more than adding transformations for show.

The challenge

The Sample Superstore dataset arrived with missing values, duplicate rows, inconsistent formatting, and no engineered features — just raw transactions with no clear story yet.

What I built

A chainable OOP pipeline (SuperstoreCleaner, FeatureEngineer) with custom exception handling — cleaning, IQR outlier detection, memory optimization, and engineered fields like Profit Margin and Discount Bucket — followed by correlation and significance testing with scipy.stats.

The result

Discount and profit are significantly negatively correlated (r = -0.22); Tables, Bookcases, and Supplies account for most of the profit loss; West/Consumer is the strongest region-segment pair at $57K profit. Delivered as 11 visualizations, a dashboard, and an automated KPI report via a custom KPIReportGenerator class.