Inspiration
Decades of learning science show that active recall and spaced repetition are the most effective study techniques — yet most students still re-read notes passively because building good practice questions is tedious. We wanted to remove that friction entirely: point AI at any material and instantly get a personalized, adaptive study experience grounded in how memory actually works.
What it does
- Generates quizzes from PDFs, pasted text, a typed topic, or a built-in AI/ML curriculum.
- Adapts difficulty per concept in real time (level up after streaks, ease off on mistakes).
- Explains mistakes instantly and offers a conversational AI tutor for follow-ups.
- Visualizes learning with a concept knowledge graph and a full analytics dashboard (improvement trends + forgetting-curve retention predictions).
- Motivates with XP, levels, badges, and streaks, plus spaced-repetition review scheduling.
- Guides students topic-by-topic through learning paths and awards a completion certificate.
- Includes multi-language support (12 languages), timer mode, progress export/import, and accessibility options (high contrast, dyslexia-friendly font, large text).
How we built it
- Frontend & app: Streamlit with a custom-themed UI (gradient branding, Plotly charts).
- AI/LLM: Google Gemini 2.0 Flash (free tier) powers six distinct features.
- Language: Python (3.9+), organized into a clean
quiz_engine/logic layer and aui/layer. - Learning science: an adaptive difficulty state machine + an Ebbinghaus forgetting-curve model.
- Deploy: one-click on Streamlit Community Cloud; secrets bridged from
st.secrets.
Challenges we ran into
- Migrating from the deprecated
google-generativeaiSDK to the newgoogle-genaiclient. - Making every AI call degrade gracefully with fallbacks so the app never crashes without a key.
- Keeping the experience cohesive as features grew — solved with shared session state and a guided-path flow that ties individual quizzes into a structured course.
Accomplishments we're proud of
- AI is the engine, not a wrapper — six meaningful AI touchpoints.
- A genuinely complete learner journey: onboarding → curriculum → adaptive quiz → tutor → graph → analytics → spaced repetition → certificate.
- Zero-cost and accessible to any student, in any of 12 languages.
What we learned
- How to translate learning-science research (active recall, spaced repetition, the forgetting curve) into concrete product mechanics.
- Prompt design for reliable structured (JSON) output and multi-language generation.
- Building a polished, stateful multi-page Streamlit app with graceful failure modes.
What's next for Adaptive Quiz Generator
- Real-time collaborative/multiplayer quiz rooms.
- Voice input/output for hands-free and accessibility use.
- Persistent accounts and cross-device sync.
Log in or sign up for Devpost to join the conversation.