-
-
Front veiw
-
Analysis of the given source that provides visualization architechture
-
MCQ conducting
-
Analysis of the MCQ
-
Anaysis of the MCQ
-
High Yield from the source topic to memorize easily
Inspiration
College students often struggle with syllabus overload right before exams. Traditional study approaches involve passively reading through 50+ page lecture decks or reading lists without knowing where knowledge gaps lie. We wanted to build an intelligent academic co-pilot that shifts revision from passive skimming to active recall, visual hierarchy retention, and targeted weakness diagnosis.
What it does
PrepPilot AI transforms unstructured academic material into an actionable revision pipeline in seconds:
- Visual Architecture Mindmaps: Automatically converts syllabus chapters into hierarchical, interconnected Mermaid mental models to visually memorize multi-unit relationships.
- Targeted 10-Question Diagnostic Engine: Generates high-yield multiple-choice questions dynamically tagged with underlying core concepts.
- Knowledge Gap Detection Deck: Instantly isolates incorrect responses and provides a single-line conceptual memory fix rather than generic boilerplate advice.
- High-Yield Takeaways: Condenses essential takeaways into clear, scannable glassmorphic cards.
How we built it
- Core Orchestrator: Python & Streamlit for a responsive, reactive dark-glassmorphism user interface.
- Neural Reasoning Engine: Google Gemini models (
gemini-2.5-flash/gemini-3.6-flash) leveraging strict JSON structured outputs for reliable evaluation workflows. - Document Parsing:
pypdffor fast text extraction across lecture slides and syllabus notes. - Interactive Visuals: Mermaid.js embedded within Streamlit components for real-time concept diagram rendering.
- Deployment: Streamlit Community Cloud with secure server-side key management.
Challenges we faced
- Model Concurrency & Fallbacks: Handling intermittent high-demand traffic spikes required designing an automated multi-tier fallback mechanism across model endpoints.
- Deterministic Schemas: LLMs occasionally produce invalid formatting when asked to generate visual syntax alongside assessment items. Enforcing native structured JSON schemas resolved diagram and quiz rendering inconsistencies.
Accomplishments that we're proud of
- Delivering an end-to-end working system from document ingest to active diagnosis in under 10 seconds.
- Enabling visual learning through real-time Mermaid memory graphs, making hard-to-recall dependencies immediately intuitive.
What we learned
- Prompt engineering for hierarchical graph representations.
- Productionizing generative AI pipelines with robust automated retries and failovers.
What's next for PrepPilot AI
- Multi-document syllabus comparison across entire semesters.
- Spaced-repetition study reminders via email and automated calendar sync.## Inspiration
Log in or sign up for Devpost to join the conversation.