Clara Jobs

Academic job hunting means checking a dozen university boards daily, each with its own terrible search, most of it irrelevant. Clara Jobs replaces that with one ranked page that refreshes itself every morning.

How it's built

The pipeline is: scrape → SQLite → rules filter → Claude score → static page. A Python scraper pulls postdoc and fellowship listings from a set of European sources (the scraper base class makes new sources drop-in). A rules filter throws out the obvious noise cheaply, then Claude Haiku scores what survives against a specific researcher profile — field, location, seniority — so the feed is ranked by fit, not recency.

The deployment trick is the same one this site uses for book highlights: a GitHub Action runs the ingest at 06:00 UTC, commits the updated database back to the repo, and that commit triggers the Vercel rebuild. The Next.js front end reads the database at build time, so the served site is fully static — no server, no API costs at read time, and the whole history of every day's feed lives in git.