Recall — Discover What You Don't Know. Master What Matters.

Inspiration

Every student has experienced the frustration of spending hours studying, feeling prepared, and then struggling on an exam. The problem isn't always a lack of effort. Often, students simply don't know what they don't understand.

Traditional studying relies heavily on rereading notes, memorizing information, and repeating practice problems. Even modern AI study tools tend to focus on generating summaries, flashcards, and answers rather than identifying the deeper misunderstandings preventing students from making progress.

We wanted to change that.

What if an AI-powered learning platform could go beyond explaining concepts and actually help students understand why they're struggling?

That question became the foundation of Recall, a personalized learning platform designed to make studying more intelligent, efficient, and meaningful.

Our vision is simple: Students shouldn't have to guess what they need to study.

What It Does

Recall transforms traditional study materials into an interactive, AI-assisted learning experience centered around understanding rather than memorization.

The platform is designed to help students uncover knowledge gaps, recognize connections between concepts, and focus their efforts on the material that matters most.

AI-Powered Study Material Processing

Students can enter their notes and learning materials into Recall, which analyzes the information and organizes key concepts into structured study content.

Rather than requiring students to manually sort through large amounts of information, Recall helps turn their material into a foundation for focused learning.

Intelligent Knowledge Mapping

Recall explores relationships between concepts and their underlying prerequisites.

For example, a student struggling with quadratic equations might actually have a weaker understanding of factoring or algebraic manipulation.

By visualizing connections between topics, Recall helps students recognize how foundational knowledge influences more advanced understanding.

Diagnostic Questions and Concept Analysis

Recall generates practice questions to help students evaluate their understanding.

Instead of treating every incorrect answer as an isolated mistake, the platform aims to connect learning difficulties to concepts that may require additional review.

This helps shift studying from passive repetition toward active problem-solving.

Personalized Learning Feedback

Recall is designed to provide targeted feedback that helps students identify weaknesses and revisit relevant concepts.

Its goal is not simply to give students the correct answer, but to help them understand the reasoning needed to arrive at that answer independently.

Privacy-Conscious AI Architecture

An important part of Recall's development has been exploring client-side artificial intelligence.

By experimenting with models that run directly in the user's browser, Recall investigates ways to reduce dependence on external AI services and offer greater control over how educational information is processed.

Together, these capabilities support a central mission:

Transform studying from memorizing information into understanding how knowledge connects.

How We Built It

Recall was developed as a modern web application using a combination of frontend engineering, AI integration, and learning-focused product design.

Our primary technologies include:

• Next.js: Application architecture and web framework.
• React: Interactive interfaces and reusable components.
• TypeScript: Type-safe application logic and maintainable code.
• Tailwind CSS: Responsive layouts and visual design.
• Transformers.js: Experimentation with browser-based language-model inference.
• ONNX Runtime Web: Supporting local machine-learning execution.
• WebGPU and WebAssembly: Exploring different browser inference backends.
• Qwen-based language models: Experimentation with local AI-powered text processing.

One of our main technical objectives was to build an educational application where AI supports a complete learning workflow rather than functioning as a standalone chatbot.

We focused on connecting note processing, concept organization, question generation, and personalized feedback into a cohesive experience.

We also explored alternative processing paths to improve reliability when local language-model execution was unavailable or unsupported.

Throughout development, our approach emphasized usability, modular application structure, and the potential for more advanced personalization.

Challenges We Ran Into

1. Running AI Directly in the Browser

One of our most significant challenges was integrating local AI inference into a browser-based application.

Unlike cloud-hosted models, browser-based models depend heavily on the user's available memory, graphics hardware, and browser compatibility.

We encountered challenges involving WebGPU buffer management, model initialization, inference performance, and browser responsiveness.

These issues made it clear that advanced AI features must be designed around real-world device limitations.

2. Integrating Machine Learning With Next.js

Combining machine-learning runtimes with a modern JavaScript framework introduced unexpected technical problems.

Some dependencies included platform-specific native components that could not be directly bundled for browser execution.

Investigating these errors helped us better understand the differences between server-side and client-side environments and the importance of selecting compatible packages.

3. Designing Meaningful Learning Experiences

Generating summaries or questions is relatively straightforward. Identifying why someone misunderstands a concept is much harder.

We had to think about how concepts relate to one another, how prerequisite knowledge influences performance, and how feedback could become more actionable.

This challenge shaped Recall's central product philosophy: AI should help students strengthen understanding, not simply automate schoolwork.

4. Balancing Innovation and Reliability

We discovered that ambitious features provide little value when they cannot run consistently.

This motivated us to explore fallback approaches, improve error handling, and consider how core learning functionality could remain useful even without successful local AI initialization.

Accomplishments That We're Proud Of

Our biggest accomplishment is developing Recall around a clearly defined educational problem instead of building another generic AI chatbot.

We're particularly proud of:

• Creating a concept-centered approach to AI-assisted studying.
• Connecting multiple stages of the learning process into one application design.
• Exploring browser-based AI as an alternative to traditional cloud-only architectures.
• Tackling real engineering problems involving model compatibility, inference performance, and application stability.
• Designing an extensible foundation for more advanced personalized learning.
• Turning an idea about educational improvement into a tangible software project.

Recall represents our effort to make AI educational tools more purposeful, accessible, and focused on genuine learning.

What We Learned

Building Recall taught us that creating meaningful AI applications requires more than integrating a powerful language model.

We developed a deeper understanding of modern web architecture, machine-learning integration, browser performance, asynchronous workflows, and robust error handling.

We also learned the importance of translating an ambitious product vision into smaller, testable features.

Most importantly, this project changed how we think about AI in education.

Rather than asking how AI can provide answers faster, we began asking how it can help students develop stronger understanding and greater independence.

The most effective educational technology should not replace thinking. It should support it.

What's Next for Recall

Recall is an evolving platform, and our long-term vision extends beyond its initial prototype.

Adaptive Learning Paths

We plan to create personalized study sequences that adjust as students demonstrate mastery of different concepts.

More Advanced Knowledge-Gap Detection

Future versions could analyze patterns across multiple responses to identify deeper prerequisite misunderstandings and suggest targeted review.

Multimodal Learning

We aim to expand support for different educational formats, including images, PDFs, handwritten notes, diagrams, and audio-based learning materials.

Intelligent Spaced Repetition

Recall could introduce automatically scheduled review sessions designed to reinforce knowledge over time.

Progress Analytics

We envision interactive dashboards that visualize student progress, concept mastery, and areas requiring additional attention.

More Efficient Local AI

We plan to investigate smaller, optimized AI models and improved inference architectures to support a wider range of devices.

Final Vision

Education shouldn't revolve around how much information a student can memorize. It should focus on how deeply they understand the concepts they're learning.

We envision Recall becoming an intelligent learning companion that helps students identify misconceptions, connect ideas, build confidence in their understanding, and take greater control of their education.

Recall isn't just about remembering more. It's about understanding better.

Discover what you don't know. Master what matters.

Built With

• html5
• javascript
• next.js
• onnx-runtime
• qwen2.5
• react
• tailwindcss
• transformers.js
• typescript
• webassembly
• webgpu