Match UCSD undergraduates to research labs and prepare them to approach a PI.

Built for the Claude Builders Club @ UCSD Hackathon, May 9 2026.

https://hackathon-two-fawn.vercel.app

https://docs.google.com/document/d/1wfO-RJdFuQL4-0yVVoo3SLtxYzg2Ft2XkiWx8QX0QLY/edit?usp=sharing

• Matches student interests against 39 hand-verified UCSD labs across Bioengineering and Physics. Vague input gets clarifying questions to guide towards labs.
• Generates a prep brief using the lab's actual recent papers and the UCSD course catalog. Foundations are tied to specific paper claims; course recommendations are restricted to courses that actually exist in the catalog.
• Runs a 2-question mock interview calibrated to the student's stated background and the lab's recent papers
• Drafts a cold email to the lab PI reflecting the student's abilities based on their strengths during the mock interview

• No fabricated courses. The model can only cite courses present in the catalog.
• No invented citations. Papers are pulled from Semantic Scholar at runtime.

Next.js · TypeScript · Anthropic Claude API (claude-sonnet-4-6) · Tailwind CSS · Semantic Scholar API

Single Next.js app, no backend storage — session state lives in the browser.

API routes:

• /api/match — matches student interests against lab dataset; flags vague input for clarifying questions
• /api/analyze — generates the prep brief; integrates Semantic Scholar with retry logic for transient failures (429 / 5xx); enforces catalog-only course recommendations
• /api/interview-init — generates calibrated practice questions
• /api/interview-eval — evaluates the student's answer; provides feedback, a model answer, and structured strengths/gaps
• /api/email-draft — drafts the cold email grounded in demonstrated strengths and gaps

• data/labs.json — 39 hand-verified labs (Bioengineering + Physics)
• data/courses.json — UCSD course catalog (Physics, Chemistry, Biology)

Andrey Belokovylenko · Ryan Safarha