The story behind each project: what the problem was, how I approached it, and what came out of it.
25 years of club ELO ratings, five chart types generated on demand by a plotting microservice, and a GPT-4o button that explains what each chart shows.
Instead of renting a course platform I built one: auth and storage on Supabase, checkout on Stripe, an admin panel to review student workbooks, and modules written as React components.
Six vertical landing pages, a bilingual ES/EN toggle, two regional price lists and a WhatsApp-first funnel — the business side of building websites.
What I learned shipping two tiny apps on top of third-party APIs — including what happens when the provider changes the rules and how to degrade gracefully.
How continuous feedback from the owner turned a simple catalog into a store with checkout, delivery integration and a marketing plan.
Analysts rated every item by hand. A clustering model — and an uncomfortable conversation about the number of categories — reached 85% accuracy.
Four analysts reviewed items one by one. A rotation-weighted metric and a store × item matrix cut the time spent on the task by 80%.
Three scripts that merge shipments, master data and historical ratings, then train a neural network, a decision tree and KNN to predict item ratings.
Built with two colleagues: onboarding with Firebase Auth, habit tracking, and a chatbot whose mood adapts to the user.
A solo project: enter a city and a number of days, get a route with places, timings and images — plus a collage generator for wallpapers.
Scrolling catalogs, cleaning prices and automating logins on streaming services to build datasets ready for analysis.
Pivot tables, lookups and conditional formatting wired together in VBA to replace hours of manual inventory work.
Why the site went through five iterations, what each one taught me, and why server-side rendering won in the end.