Quasar

• 

  Remember at the speed of sight :)

Inspiration

Quasar started from a problem I kept running into as a student: I could understand something perfectly while studying it, then struggle to recall it a few days later. At the same time, AI made getting explanations almost effortless, so the bottleneck no longer felt like access to information, it felt like retention. I became interested in memory techniques like visual associations, memory palaces, sound connections, spaced repetition, and active recall, and I started wondering what would happen if an app could automatically turn difficult material into something your brain actually wants to remember. That idea became Quasar: not another place to consume more information, but a memory layer that helps what you learn stay with you.

What it does

Quasar is an AI-powered learning and memory app that transforms difficult material into memorable experiences. Instead of only giving students explanations, it turns information into visual stories, sound-based connections, personalized mnemonics, and active recall exercises. Some of the current experiences include:

• Visual vocabulary and mnemonic learning
• Korean language learning that connects sound, script, imagery, and meaning
• Cell City, where cellular respiration becomes an episodic visual story
• Teach-Back, where learners explain concepts in their own words
• Why Ladder and Case Lab for deeper understanding and application
• Error Memory, which creates new memory hooks around repeated mistakes Quasar also uses spaced repetition so concepts come back before they completely fade

How I built it

Quasar is built as a cross-platform app using React Native, Expo, React, and TypeScript, with Supabase handling authentication, Postgres, storage, and Edge Functions. The main stack includes:

• Claude through Anthropic’s API for mnemonic generation, quizzes, feedback, and memory hooks
• FSRS for spaced repetition scheduling
• React Native Reanimated, Moti, and SVG for visual interactions
• RevenueCat for subscriptions
• Playwright for end-to-end testing The AI does not run directly from the client. Requests go through Supabase Edge Functions so API keys stay protected and responses can be validated before they are shown to the learner. I also designed the AI outputs to be structured instead of relying on unrestricted chatbot responses.

Challenges we ran into

One of the biggest challenges was realizing that something being memorable does not automatically make it useful. A strange image might stick in your head, but if it does not help you reconstruct the original word, concept, or meaning, then it has failed as a mnemonic. I also had to make sure Quasar did not slowly become another generic AI study app. While building each feature, I kept coming back to one question: does this actually help the learner retain, retrieve, or apply the information later? Other challenges included making AI outputs reliable, keeping the experience useful without constant AI access, and designing one memory system that can still work across very different subjects.

Accomplishments that we're proud of

I am most proud that Quasar grew from a simple mnemonic generator into a much broader learning system. It now combines memory creation, retrieval, understanding, application, and spaced review instead of focusing on just one part of studying. A few milestones I am especially proud of are:

• Building a working cross-platform product for web, iOS, and Android
• Creating Cell City and other visual learning experiences
• Building Teach-Back, Error Memory, and deeper reasoning tools
• Creating a secure AI pipeline rather than simply placing a chatbot inside the app I also like that AI enhances the product rather than being the entire product. The underlying learning experience still matters even without the model.

What we learned

The biggest thing I learned is that good learning design is not just about explaining something clearly. The harder question is what happens after the explanation disappears. A student can understand something but fail to recall it, remember an image but forget the meaning, or repeatedly choose the same wrong answer. Those are different problems, and they need different types of support. On the technical side, I learned a lot about building AI as a controlled part of a product through structured outputs, validation, authentication, quotas, and fallback behavior. I also realized that real personalization should go beyond difficulty. The more interesting question is how does this particular person remember best?

What's next for Quasar

Next, I want Quasar to make learning much more immersive and personal. For biology, that means interactive 3D models students can explore in their own space with their fingers, instead of only looking at diagrams. I also want users to generate custom memory episodes in different art styles, from cartoon to cinematic, depending on what helps them remember best. A built-in Quasar bot would be available anywhere in the app, even while watching reference videos, to answer questions the moment they come up. At the same time, I want to keep improving the mnemonic engine so the connections between visuals, sounds, meanings, and concepts become much stronger and more personalized.

Built With

• moti
• postgresql
• react
• supabase
• typescript