01 / Agent engineering
Browser Agent Evaluation
Improved a vision-language browser agent from 73.3% to 95.6% on a fixed 45-episode MiniWoB++ regression set. Seven paired experiments, three wins retained and four candidates cleanly reverted.
Portfolio
Agent engineering, data systems, safety-first prototypes and independent craft.
01 / Agent engineering
Improved a vision-language browser agent from 73.3% to 95.6% on a fixed 45-episode MiniWoB++ regression set. Seven paired experiments, three wins retained and four candidates cleanly reverted.
02 / Industry project
Built an AI-assisted data tool for a BCIT industry client. It imports spreadsheet data into MySQL, translates natural-language questions into SQL and returns formatted tables or raw JSON.
03 / Independent side business
I leathercraft and advertise my work online through a storefront I built and operate. The project combines handmade goods, product photography, e-commerce, hosting and practical OpenSCAD workshop tools.
Visit Ariadne LeatherI am building a media curation and suggestion product where people organize the music, films, books and games they care about, receive explainable AI-assisted suggestions and deliberately send — or “Pelt” — a pick to a friend. The current Next.js concept uses seeded sample data while the product is validated.
Built and closed an SMS intake prototype for BC electrical permits. It validates contractor and job details, prepares a draft, records receipts and deliberately blocks final submission. The archived prototype finished with 120 automated tests.