Problem
Syllabi arrive as messy PDFs, with dates, weights and deadlines buried in tables and prose. Students need them as a clean schedule, fast, without the tool doing their coursework for them.
Approach
A Next.js and React frontend talking to a FastAPI backend with MongoDB, packaged with Docker. Uploaded PDFs go through pypdf and then Gemini to pull out assessments and dates, with a regex fallback when the model call fails.
The browser's Web Speech API drives a hands-free voice onboarding flow, with academic-integrity guardrails built in.
On the frontend, I built the glowing bloom that reacts to voice input and output, the user flow starting from the loading screen, and the light, bright-coloured dashboard and UI.
Results
Built in 24 hours at GDG Hacks 3. It taught me to scope a demo-ready product under a hard clock and to keep LLM output structured enough to drive a real interface.
The extraction step raised a question worth testing properly, which became my first paper.