Earlier projects, still worth a look. Filter by topic below.
Local analytics stack replacing Google Sheets & BigQuery — GDPR compliant, vendor-free.
Visual analytics pipeline with GCP, JDBC, and live SQL queries.
Random Forest regression pipeline on live BigQuery — 3.5× accuracy improvement.
Econometric investigation with OLS and HAC-corrected errors.
Finalizing and documenting the GoExplore analytics projects.
DIY Arduino builds from scratch — a tracked breadboard robot, an ultrasonic 4WD car, the official Arduino Robot, a windmill mechanism, and a six-legged hexapod.
One ecosystem, three outcomes — a heavily upgraded mBot Ranger, a short-lived Neuron kit, and a laser cutter waiting on spare parts.
From a vintage boolean-logic kit to micro:bit, littleBits, Makeblock Neuron, and beyond — teaching kids to build and code, one snap-together module at a time.
A Meccano progression — Easy 1, Mechanical 2, Motorized 3, and two fully programmable Meccanoid robots.
A robotics progression — Useless Box, Yeti, Asuro, and the RP5/RP6v2 fleet — plus a full technical retrospective on closed-loop embedded control.
An analysis of a Brazilian e-commerce platform using SQL and Tableau. This was my first real data project — a case study using MySQL and Tableau to answer a business question for a real-world scenario.
This project started as a messy, real-world dataset from Eniac, an e-commerce electronics retailer, and turned into a deep dive on how much their discounting habit was actually costing them. The dataset was genuinely chaotic — broken price formats, duplicate rows, orders in five different states — so a large chunk of the work was just getting the data into a state I could trust.
Eniac wanted to know whether tweaking their "SHOP NOW" button — color, copy, or both — would improve click-through rate. The obvious framing is "A/B test it." But with two variables and four combinations, this is technically a 2×2 multivariate test...