Balong Supnet

The capstone centered on organizing evaluation data into a usable structure, designing a clean dashboard interface, and presenting KPIs in a way that supported non-technical stakeholders. The result was a practical reporting tool that connected data collection, backend storage, and front-end visibility into one streamlined workflow.

Monthly category counts across moment types. Day-by-day average rating movement.

The app uses text extraction for PDFs and local OCR for image invoices, then applies deterministic parsing, item classification, ingredient alias mapping, and unit conversion logic. Matched invoice rows are rolled into sandwich recipes so users can compare ingredient cost, selling price, gross profit, gross margin, and component-level changes over time.

Estimated sandwich cost is compared against the fixed sale price. Ingredient-level trends are shown with supporting invoice rows.

This project focused on preparing and analyzing credit data, selecting classification techniques, and evaluating model performance using common metrics such as accuracy, precision, recall, and confusion matrices. Beyond the technical model output, the work emphasized how data patterns can inform business decisions around risk segmentation and responsible lending.

Best-model classification outcomes across low-risk and high-risk borrowers. Feature importance ranking for the variables with the strongest influence. Model discrimination performance across the tested classifiers. Example borrower records with predicted probabilities and final classifications.