Recall — Learning Beyond Memorization

Inspiration

Students often spend hours studying without understanding why they struggle with certain concepts. Traditional tools focus on summarizing content, generating flashcards, or supplying answers, but they rarely identify the underlying misunderstandings that prevent real learning.

Recall was inspired by a simple question: What if AI could help students discover exactly what they don't understand?

We wanted to create a more intelligent study experience that connects individual mistakes to missing foundational knowledge.

What It Does

Recall is an AI-powered educational web application designed to help students study more effectively through concept analysis and personalized learning support.

Its core capabilities include:

AI-Assisted Note Processing: Organizes study notes into useful learning material.
Knowledge Mapping: Visualizes relationships between concepts and foundational knowledge.
Diagnostic Questions: Helps students test their understanding.
Personalized Feedback: Supports identifying concepts that require further review.
Privacy-Conscious AI: Explores client-side AI processing to reduce reliance on external inference services.

Recall's broader vision is to support learning across multiple formats, including text, images, and documents, allowing students to connect information from different types of educational material.

How We Built It

Recall uses Next.js, React, TypeScript, and Tailwind CSS for the web application.

We explored local language-model inference with Transformers.js, ONNX Runtime Web, and browser-supported acceleration technologies such as WebGPU and WebAssembly.

The application is structured around a learning workflow that connects study material, concept extraction, question generation, and feedback.

Our technical focus has included balancing advanced AI capabilities with browser compatibility and reliability.

Challenges We Ran Into

One of the most significant engineering challenges was running AI inference directly inside a browser.

We encountered issues involving WebGPU memory, ONNX Runtime compatibility, long model initialization times, and browser responsiveness.

Another major challenge involved making AI-generated study material educationally meaningful rather than simply producing generic questions.

These challenges demonstrated the importance of reliable application architecture, useful fallback experiences, and concept-aware learning design.

Accomplishments We're Proud Of

We're proud of developing Recall around a specific educational challenge: helping students recognize gaps in their understanding.

The project combines learning-oriented interface design with AI experimentation and a concept-based approach to education.

Rather than treating AI as a simple answer generator, Recall explores how it can help students become more independent learners.

What We Learned

Building Recall taught us how modern web technologies and AI inference systems can be combined to create educational applications.

We gained experience with frontend architecture, TypeScript, browser-based machine learning, performance optimization, debugging, and AI-assisted software development.

We also learned that useful educational AI needs to be reliable, understandable, and focused on helping students learn.