I built a "what do I need on my final" calculator with AI auto-fill (Next.js 16 + Gemini)

A developer built GradeHQ, a final grade calculator using Next.js 16 and Gemini AI auto-fill. The tool lets students input weighted course categories and scores, then solves for the required final exam score to achieve a target letter grade. It also supports uploading syllabus PDFs or grade screenshots for automatic extraction of categories, weights, and scores via Gemini 2.5 Flash, with all data stored locally in the browser.

Every finals season I'd end up doing the same algebra on a notes app: "I have a 91% in homework 20% , 84% on the midterm 30% ... what do I need on the final 50% to get an A-?" So I built GradeHQ to do it for me — and then kept adding to it until it became a real side project. Live: gradehq.vercel.app Code: github.com/yashmitb/gradecalc MIT What it does Add your courses with weighted categories homework, midterms, final, etc. Enter scores as you get them back Pick a target letter grade and it solves for exactly what you need on whatever's left Optional AI auto-fill: drop in a syllabus PDF and/or a screenshot or PDF of your Canvas grades, and it extracts your categories, weights, and scores for you Everything is local-first — courses are stored in localStorage, nothing's saved server-side except the file you choose to auto-fill, which is sent once to Gemini and discarded. Stack Next.js 16 App Router, Turbopack TypeScript Tailwind v4 Framer Motion for the animations/transitions Gemini 2.5 Flash for the auto-fill parsing No backend database — just one API route /api/parse that proxies to Gemini The auto-fill flow This was the fun part. The /api/parse route takes the uploaded files PDF/PNG/JPG, up to 4 , sends them to Gemini with a prompt asking it to extract { name, categories: { name, weight, score } , flags: }. A few things I had to handle: Uncertain extractions shouldn't be silently guessed. If Gemini can't confidently read a score or weight, it returns a flags array explaining what it left blank and why. The UI shows these to the user before importing, so they can fill them in themselves instead of getting a wrong number with no explanation. Shared vs. personal API keys. The app ships with an optional shared/free Gemini key server-side env var so most users don't need to do any setup. If that key hits its daily quota, the API returns a server-busy code and the UI lets the user paste their own free key inline to retry — saved only in their browser. Making a 10-20 second wait feel okay. Gemini calls on multi-page PDFs can take a while. Instead of a static "Loading...", the extract button cycles through short status messages "Reading files…", "Scanning syllabus…", "Matching scores…", etc. every ~1.3s, tailored to whether you uploaded a PDF, an image, or both. "Almost done…" only shows up after the cycle has looped through twice — so it doesn't lie to you on a request that just started. What's next More grading scale presets currently a standard A+ through F scale Better handling of edge cases in syllabus formats Probably a dark/light theme toggle since right now it's dark-only Repo's open source and PRs are welcome — especially around the parsing prompt, since LMS export formats vary a ton between schools. Feedback appreciated