Projects · Selected Work
Technology Investment Business Case
Group · Investment case · Budgeting · Risk analysis
An e-commerce company had customer data it couldn’t use, with browsing, CRM, and purchases trapped in three disconnected systems, and had to decide whether fixing it was worth a roughly $275K first-year investment. My piece was the financial and planning backbone of that decision: comparing the options on cost, timing, and execution risk, then building the line-item budget, an 18-month phased rollout, a risk register with mitigations, and the assumptions behind the investment case, plus the dimensional data model underneath the reporting layer.
Read the business case →
Chicago Airbnb Price Analysis
Solo · R · Multiple regression
If you’re pricing an Airbnb in Chicago, two things actually move the needle: what kind of space it is and where it sits. The review count hosts agonize over barely matters. I got there by building up five regression models across ~6,300 listings, layering in location, reviews, and availability, then log-transforming price to handle the skew from luxury outliers. Room type and neighborhood were the real levers. The point isn’t the R². It’s the question an investor asks of any asset, what actually drives its price, answered with evidence, and being honest about the ~60% the data can’t explain.
Read the full analysis →
Stockscribes: Pharmacy Stock Ordering Platform
Prototype
Solo · Python · SQLite · JavaScript
A family-owned pharmacy in Lagos, Nigeria was ordering all of its stock by phone, calling each wholesaler in turn to ask what was available and at what price. There was no way to compare suppliers, no visibility of expiry dates before delivery, and no record of what had been agreed. Stockscribes replaces that with a single shared catalogue: pharmacy staff search by product or generic name and see every supplier’s price, stock and expiry side by side, with short-dated batches flagged since they are priced lower. The decision that shaped the architecture was making suppliers users of the system rather than records in it. They sign in and maintain their own listings, which costs them less time than answering the same calls and gives the pharmacy live pricing without manual upkeep. Built end to end (schema, backend, interface, deployment) in Python with no frameworks, using AI as a build partner. A working prototype with sample data, not yet in daily use.
Try the live demo →
Renewable Energy & Weather Analysis
Group · R · Regularized regression
Renewable energy output turns out to be genuinely predictable from weather, and solar conditions dominate everything else. On a dataset of 100,000+ observations, the response was heavily skewed and the weather variables were tangled together, so I square-root-transformed the response and used regularized regression (Ridge, LASSO, Elastic Net), which is built for correlated predictors like these. Tuned with cross-validation, Elastic Net predicted about 85% of the variation in energy change, with solar irradiance the standout driver, the kind of predictability that makes an energy asset easier to plan around and to finance.
Read the full study →