LastMinuteKU

Its 4 AM and just 5 hours before the exam. I need to pass. Such Last Minute Study Assistant. Lets Pass

Comment

Inspiration

LastMinuteKU was inspired by the real exam panic most students face a few hours before finals. We wanted to build something that turns “I have too much to study” into a fast, structured, and personalized plan.
Instead of forcing students to manually sort through folders of notes, books, and past papers, we designed a tool that can understand mixed academic material and instantly generate the exact study assets needed to pass.

What it does

LastMinuteKU is an AI-powered exam preparation web app that connects to Google Drive study folders and helps students go from chaos to focused revision.

It lets users:

  • Browse and preview selected PDFs from Drive
  • Generate exam-focused study guides
  • Generate interactive quizzes with instant correct/wrong feedback
  • Generate presentation slides in a slide-by-slide viewer
  • Generate flashcards in a flip-card viewer
  • Export outputs in LaTeX format for external viewing/printing
  • Toggle visual generation on/off for faster or richer outputs
  • Use a landing-to-study flow optimized for quick start under pressure

It also adapts to mixed source types:

  • Past papers for exam patterns
  • Notes/slides for instructor emphasis
  • Textbooks for conceptual depth

How we built it

We built LastMinuteKU as a full-stack web app using:

  • Node.js + TypeScript + Express for backend APIs
  • Google Drive API for folder/file browsing and PDF access
  • Gemini API for generating study outputs
  • Custom prompt orchestration tailored to past papers, notes, and textbooks
  • Vanilla frontend (HTML/CSS/JS) with feature-specific UX modules

Technical highlights:

  • Recursive Drive tree with scoped folder restrictions
  • PDF preview route and text extraction pipeline
  • Structured generation endpoints for guide/quiz/slides/flashcards
  • Interactive quiz engine with answer logic and explanations
  • Slide/card navigators for presentations and flashcards
  • LaTeX generation/export + viewer integration
  • Retry/fallback logic for model availability and rate-limit resilience
  • Cache + lazy rendering + port self-healing startup script

Challenges we ran into

  • Model/availability issues: some requested model names were unavailable or overloaded, causing 404/503 responses.
  • Quota/rate-limit pressure: generation across multiple features could trigger 429/503 spikes.
  • Push protection on GitHub: secret scanning blocked deploy pushes due to credential files in history.
  • Large payload UX issues: LaTeX viewer initially failed with URL length limits (414 errors).
  • Mixed-source consistency: combining past papers, notes, and books while keeping outputs exam-relevant required careful prompt engineering.
  • Performance with large Drive trees: initial rendering and repeated tree scans needed optimization.

Accomplishments that we're proud of

  • Built an end-to-end usable product for exam prep, not just a demo API.
  • Delivered a polished student journey: landing page -> start studying -> feature workbench.
  • Made quiz outputs truly interactive (instant visual correctness + explanation).
  • Added dedicated viewers for slides and flashcards for practical study flow.
  • Created source-adaptive generation logic for real university content structure.
  • Implemented robust resilience (fallbacks, retries, clearer errors).
  • Successfully cleaned git history and shipped while passing secret protection rules.
  • Kept the app fast enough for hackathon context with caching and lazy UI rendering.

What we learned

  • AI products need strong UX structure as much as strong models; raw output alone is not enough.
  • Prompt quality improves dramatically when we encode source-type priorities (past papers vs notes vs books).
  • Reliability patterns (fallbacks/retries/rate-limit-aware spacing) are critical for production-like behavior.
  • Security and delivery are part of product velocity; secrets and push policies can block shipping if ignored.
  • Feature-specific interfaces (quiz runner, slide viewer, flashcard viewer) increase real user value compared to a single generic output panel.
  • Simplicity wins in hackathons: focused flows and deterministic outputs beat over-complex architectures.

What's next for LastMinuteKU

  • Add a dedicated time-to-exam study planner (e.g., 2h/4h/8h) with adaptive revision schedules.
  • Add chapter/topic filtering before generation.
  • Add downloadable formats (PDF-ready notes, PPT-ready slide export, Anki/CSV flashcard export).
  • Improve collaborative functionality (shareable study sessions/links).
  • Introduce lightweight analytics (time spent, weak-topic tracking, revision heatmaps).
  • Add optional user accounts for saving progress across sessions.
  • Deploy a production version with proper secrets management and monitoring.
  • Expand support for additional course folders and broader university cohorts.
Share this project:

Updates