Nuclear has a dev blog if you want to learn more about how we structured our weeks!! https://dev.nuclearapp.ca/

Inspiration

We noticed that while note-taking tools are abundant, none truly understand your notes. Especially for students and lifelong learners, organising ideas, remembering key points, and truly grasping the material remains a challenge. We wanted to build an AI-powered platform that doesn’t just help you take notes, but helps you think, revise, and master the material. That’s how Nuclear was born.

What it does

Nuclear is an AI-powered note-taking assistant that turns raw thoughts into structured knowledge. Users can:

Transcribe and summarise voice notes using Perplexity Sonar
Fill in the blanks and generate questions, Duolingo-style
Extract key concepts and definitions
- Generate examples that explain abstract topics
- Auto-tag and organise notes into folders
- Engage with notes interactively using a natural language interface

We wanted to create a space where writing is just the beginning—and learning continues, actively and intelligently.

How we built it

Nuclear is built with:

Frontend: Next.js + Tailwind CSS + Our own custom UI library (based on NyxbUI)
Backend: Prisma + Supabase Postgres

AI integration: Perplexity Sonar API powers summarisation, concept extraction, and interactive features

Voice recording: AssemblyAI API with browser compatibility considerations

OCR: AWS Textract for text recognition in images and PDFs.

Storage: Supabase object storage for short-lived audio files, Supabase Storage bucket

We fine-tuned prompt engineering pipelines to balance speed and depth across different learning modes.

Challenges we ran into

Sonar Rate Limits: We had to optimize API usage, balancing model calls with user flow.

Prompt design: Making prompt chains that didn’t just summarise, but teach required iterative design.

Browser limitations: Voice recording compatibility, especially with Firefox, required fallback logic.

UX Design: Creating a fun but robust UX for different learning styles wasn’t easy—we tested a lot.

Accomplishments that we're proud of

A working MVP with all major features

A smooth and clean user interface that's actually fun to use

Built a scalable backend with Supabase + Prisma

Got positive feedback from real student testers on how useful and engaging the app felt

What we learned

Prompt engineering is UX: How your prompts are shaped directly changes the user experience.

APIs like Sonar can feel magical—but only with structure: Wrapping them in good UX, fallbacks, and tight feedback loops is key.

Even minimal voice input creates high engagement if you respond to users fast and smartly.

AI doesn't replace learning—it supercharges it: Our goal became clearer: not to replace teachers or textbooks, but to augment learning in a personalized way.

What's next for Nuclear

Launch a beta waitlist and begin onboarding early users

Add multilingual support, especially for international learners

Implement adaptive revision plans using spaced repetition

Seek funding or grants to sustain development and compensate our team

Expand our AI toolkit to include diagrams, timelines, and mind-maps

We're incredibly excited to continue building Nuclear as a smart companion for every curious mind.