Every finals season, students end up doing the same tedious algebra on their note apps: "My homework is worth 20% and I have a 91%, midterms are 30% and I got 84%... what do I need on the 50% final to get an A-?" To automate this, developer yashmitb built GradeHQ, expanding it into a feature-rich, open-source side project.
The application allows users to input courses with custom-weighted categories and current grades. After selecting a target letter grade, GradeHQ solves for the exact score needed on the remaining assignments. Crucially, it features an AI auto-fill flow: users can drop in a syllabus PDF or a screenshot of their Canvas grades, and the system automatically extracts categories, weights, and scores.
The tech stack is highly modern, utilizing Next.js 16 (leveraging App Router and Turbopack), TypeScript, Tailwind v4, and Framer Motion for smooth animations. For the AI-powered backend, the developer chose Gemini 2.5 Flash. Adhering to a #local-first philosophy, all course data is stored in the browser's localStorage, with files temporarily proxied to #Gemini via the /api/parse endpoint and immediately discarded.
Handling the AI parsing flow presented interesting challenges. To prevent inaccurate data, if Gemini 2.5 Flash is uncertain about a specific extraction, it returns a descriptive flags array rather than guessing, allowing the UI to prompt users for manual correction. Additionally, the app features an integrated shared API key fallback, letting users paste their own free Gemini keys locally if rate limits are hit. The developer also implemented a dynamic loading UI that cycles through contextual status messages to keep users engaged during the 10-20 second file-processing window.
Future plans include adding more grading scale presets, robust edge-case handling for varying syllabus formats, and light/dark mode toggling. The repository is open-source and open to contributions.
[AgentUpdate Depth Analysis] GradeHQ exemplifies a highly practical blueprint for the future of AI-driven vertical applications: combining lightweight frontends with powerful micro-LLMs like Gemini 2.5 Flash for seamless, local-first workflows. In the AI Agent ecosystem, unstructured data ingestion (such as parsing diverse academic syllabi and inconsistent LMS screenshots) remains a prominent bottleneck. GradeHQ successfully circumvents this by using LLMs as structured extraction engines, paired with a robust "Human-in-the-loop" verification system using uncertainty flags. This pattern of combining local storage, proxy-based AI parsing, and intuitive user fallback mechanisms represents a scalable approach to building friction-free, specialized AI agents that respect privacy while vastly outperforming traditional manual-entry utilities.
