Ursearch

LOG · WHAT IT IS AND WHAT IT DOES

A platform that matches students with research positions at UW-Health. A multi-layer weighted attribute algorithm scores each applicant against a position; in a trial with about 30 applicants, its top 3 were the same people four rounds of interviews picked. It was presented at IADR 2024.

WHY

The underlying matching algorithm came about when I met someone at a party who said they were looking to get matched with students for research more easily, and I said I could make that algorithm. I was 19 at the time.

WHAT I WORKED ON

I researched the leading matching platforms' algorithms to learn what goes on under the bonnet, and created a multi-layer weighted attribute system. We needed to hire more engineers for the frontend, so we put out an application and got about 30 applicants. We ran them through the algorithm and saved the results without looking at them, then did 2 rounds of technical interviews and 2 rounds of personal interviews, and selected 3 top candidates. Comparing that selection against the algorithm's results showed 93% accuracy: the top candidates we picked through traditional interviews were the same as the algorithm's top 3 matches. The algorithm was in Python, the backend in Django, and the frontend in React/Next.js.

WHAT CAME OF IT

The results were so impressive that serious talk about protecting the algorithm's IP came up at the time. Despite the interest, it got caught up in university bureaucracy and lost in the abyss, until it was eventually cleared by all parties and left unclaimed. Then in 2026 it was picked up by UMich: I was brought back to lead the development of a new and improved version, which became Scholara.