problem · Redesign a global coffee producer's distribution network by siting a second warehouse to minimize annual transport and handling costs across 17,600 tonnes of demand.
L1
L2
L3
L4
L5
L6
Built a linear programming model in Excel Solver minimizing LTL + FTL + handling costs across 6 candidate warehouse locations vs. a central-only baseline.
Modeled the 40% FTL discount on inter-warehouse transfers and per-tonne LTL rates across 6 distribution-center customers; solved 7 scenarios with sensitivity reports.
Identified Location 4 as optimal: total annual cost dropped from €10.10M to €8.55M, a €1.55M (~15.4%) reduction.
Routed Customer 4's 5,500-tonne demand through the new satellite (FTL backbone); kept the other 5 customers on the central warehouse.
Power BIBIComplianceDAXStar SchemaAnomaly Detection
Workforce BI: JW Marriott San Antonio
problem · Bridge two disconnected workforce systems (WhenToWork scheduling + Kronos punch clock) at a luxury hotel to detect compliance risk before payroll close.
Built a 5-page Power BI dashboard plus an executive narrative on a star-schema data model spanning 8.5 months and 18 bi-weekly pay periods.
Implemented DAX measures for 7-minute punch rounding, weekly overtime (Sunday to Saturday, 40-hour threshold), missed-punch detection, and night/marathon-shift anomalies.
Built compliance trackers for the 19-hour weekly cap and 1,000-hour cumulative-tenure ceiling, flagging part-time student workers approaching benefit-eligibility thresholds before a breach.
Layered anomaly detection for rounding-rule gaming (consistent 6-7 minute punches) and impersonation (identical-time punches across consecutive shifts).
problem · Help a Brazilian e-commerce marketplace cut client acquisition cost while protecting merchant revenue as post-pandemic tailwinds reversed.
Paid
Direct
Social
Display
Email
Built a Power BI dashboard from a MySQL extract of 8,000 leads and 842 converted merchants, mapping a 10.5% overall conversion rate across acquisition channels.
Surfaced Paid Search (12.3%) and Direct Traffic (11.2%) as the strongest channels vs Social (5.6%) and Email (3.0%); recommended reallocating 30% of underperforming spend to lift conversion to 13%.
Quantified revenue concentration risk: construction tools generated R$50.7M (~80% of merchant revenue), with manufacturers outperforming resellers 5 to 1.
problem · Cut delivery cost and CO₂ for a furniture retailer by solving the daily Vehicle Routing Problem and feeding optimized routes into a live Power BI dashboard.
Built a Python pipeline using Google OR-Tools to solve a capacity-constrained Vehicle Routing Problem on customer orders pulled from SharePoint Excel.
Engineered a bin-packing scheduler with time-window feasibility, splitting oversized orders across multiple delivery nodes.
Computed cost and CO₂ for each route vs. a naive baseline and exported 4 CSVs powering a single-refresh Power BI dashboard.
Designed for live disruption testing: edit order status in SharePoint, hit Refresh in Power BI, and the optimizer reruns end to end.
problem · Decode why some Airbnb listings outperform others by mining 3,985 listings across 12 markets and isolating which content signals (not pricing) drive guest satisfaction.
Ran TF-IDF, bigram extraction, and Bing-lexicon sentiment scoring across listing descriptions and reviews to surface the vocabulary that separates top-rated from low-rated stays.
Showed that content quality, not tone, drives review scores: high-rated listings used markedly different vocabulary while sentiment polarity was nearly flat across rating groups.
Compared 12 markets and room types side by side; entire-home listings carried both higher prices and stronger review consistency.
Delivered 6 chart-ready visuals and 7 analytical CSVs as a reproducible R pipeline.