Academic Matchmaker

Choosing a PhD or postdoc advisor is one of the biggest decisions in an academic career, and it is still done by hand. Students read faculty pages one university at a time, guess at fit from paper titles, and cold-email people whose research moved on years ago.

The Academic Matchmaker turns that into a search problem. You drop in a CV, a paper or a Google Scholar screenshot, and it ranks researchers across the top 200 universities by how closely their recent work matches yours. Two constraints shaped every decision:

• Nothing you upload leaves your device. Text extraction, the language model and the matching all run in the browser.
• No servers. The site is static files on GitHub Pages. The data is rebuilt by GitHub Actions and served from a Hugging Face dataset.

Those two constraints are what make this build interesting. Fitting 181,568 researchers into something a browser can search in a second or two took a series of storage and ranking decisions. They are worth writing down because the same moves apply to any retrieval system with a tight budget.

The three parts

Part 1 covers the problem and the architecture. Part 2 is the data pipeline and storage: how 200 universities of OpenAlex data are ingested under a free API budget and squeezed from about 280 MB of vectors down to under 9 MB. Part 3 is matching: the small model in the browser, coarse-to-fine vector search, why pure semantic matching was not enough, and the concept tags and laureate test set that fixed it.