Scaffold Ai

Inspiration

As students, we are constantly drowning in information but starving for understanding. We all know the struggle: you stare at a 500-page textbook PDF or messy lecture notes, read them for hours, and feel like you've studied. But when the exam comes, you realize you only had the illusion of competence.

We realized that current AI tools are part of the problem. Standard chatbots just give you the answer, which is passive and doesn't help retention. We wanted to build something that uses Instructional Scaffolding—a method where the tool supports your learning process rather than replacing it. We built Scaffold Ai to turn the chaos of raw course materials into a structured path to mastery.

What it does

Scaffold Ai is an active learning platform that transforms static documents into interactive study experiences.

Intelligent Ingestion: Users upload their raw PDF textbooks or notes.
Modular Breakdown: Instead of a wall of text, the system breaks the content down into simplified, digestible learning modules.
Auto-Generated Slides: The AI extracts key concepts to create visual summary slides, perfect for quick review before class.
Active Recall Quizzes: At the end of every module, the system generates a quiz to test your knowledge immediately. If you get it wrong, it doesn't just tell you the answer—it points you back to the source material.
Context-Aware Chat: We include a chatbot, but it's grounded in your specific documents. It acts as a tutor to clarify difficult concepts from the slides or modules.

How we built it

We built Scaffold Ai using a Retrieval-Augmented Generation (RAG) architecture to ensure accuracy.

Frontend: We used Next.js to create a clean, distraction-free student interface.
Backend: Powered by ** Python/FastAPI ** to handle file processing.
AI & Logic: We utilized Gemini for the reasoning engine.
The Pipeline: When a PDF is uploaded, we use a parser to extract text and structure. This data is chunked and stored in a vector database. When a user starts a module, the AI retrieves relevant chunks to generate the summary slides and corresponding quiz questions dynamically.

Challenges we ran into

PDF Parsing: PDFs are notoriously difficult to scrape cleanly. Handling multi-column layouts and random formatting in textbooks was a significant hurdle.
Hallucinations vs. Grounding: We had to spend a lot of time on prompt engineering to ensure the quizzes were based strictly on the uploaded PDF, preventing the AI from asking questions about things the user hasn't learned yet.
Slide Formatting: Teaching a text-based LLM to output content that looks good in a "Slide" format required creative prompting and structural constraints.

Accomplishments that we're proud of

The "Teacher" Persona: We successfully tuned the AI to feel like a supportive tutor rather than a robot search engine.
The Quiz Logic: Getting the system to generate relevant, non-trivial questions that actually test understanding was a big win.
Integrated Workflow: Successfully combining the Chatbot, Slides, and Quizzes into one seamless UI so it feels like a cohesive app, not just three separate features glued together.

What we learned

Active Recall is King: We learned deeply about educational psychology. Simply reading text is the least effective way to learn; testing yourself is the best.
Context is Key: The quality of the AI's output is entirely dependent on how well we clean and chunk the input PDF.
The Student Experience: Building for ourselves showed us that speed and simplicity matter most. Students don't want to configure settings; they want to upload and start learning.

What's next for Scaffold Ai

Handwritten Notes: Expanding our ingestion engine to read images of handwritten notes using OCR.
Gamification: Adding streaks, XP, and leaderboards to keep motivation high during midterms.
Spaced Repetition: Implementing an algorithm to surface old flashcards/quizzes right before you are about to forget them.
Audio Mode: Generating "Podcast" summaries of the modules so students can learn while commuting.