Intern Finder

A web app for finding internships across every major, with a built-in application tracker. It installs on phones and laptops like a native app (PWA).

Live: https://intern-finder-xi.vercel.app/

What it does

• Search thousands of open internships with company logos, ratings and AI-written summaries of each posting
• Resume matching: upload a resume and an AI model ranks postings by fit
• Application tracker that works signed out (saved in the browser) or signed in (synced across devices)
• Gmail inbox tab that reads application emails, read-only and entirely in the browser (see the note below on access)
• Installable and offline-friendly with a service worker and web app manifest

Note on Gmail access

The Gmail inbox feature is not open to everyone. Google requires an app that reads Gmail to pass a restricted-scope verification review, and this project has not gone through it. The Google app is in Testing mode, so only people I add as approved test users (up to 100) can connect their Gmail. Everyone else can still use search, resume matching and the application tracker; only the Gmail tab is unavailable. If you want to try the Gmail tab, send me the Google account email you'd like added.

How it's built

Layer	Tech
Front end	HTML, CSS and vanilla JavaScript, shipped as a PWA
Data pipeline	Python scripts that pull from public internship lists, company career sites and program pages, then rebuild data.json
Automation	GitHub Actions runs the pipeline every hour and commits new listings; Vercel redeploys automatically
Backend	Vercel serverless function (Node.js) for resume matching, calling Gemini or Claude. API keys stay on the server, with per-IP rate limiting
Accounts	Supabase (Postgres + email/Google sign-in) with row-level security so people only see their own data
Gmail	Google OAuth with the read-only Gmail scope, run client-side
Logos	Logo.dev

Design decisions

• Privacy first. Gmail access is read-only and runs client-side, so nothing from a user's inbox reaches the server. Resumes are sent only to the matching function and API keys never reach the browser.
• No server to maintain for listings. The pipeline writes a static data.json, so the site is fast and cheap to host.
• Model fallback. The matching function tries several models in order and falls through on overload, quota or retirement errors, so a single model outage doesn't break the feature.

Repo layout

intern-finder/
  public/      the website (index.html, app.js, data.json, PWA files)
  api/         serverless resume-matching function
  pipeline/    scripts that rebuild the listings
  supabase/    database schema for accounts
.github/       hourly refresh workflow

License

This repository is view only. All rights reserved; see LICENSE. Please do not copy, deploy or reuse the code without permission.

Author

Ethan, Computer Engineering at the University of Miami.