Intercortex

Word -> AI Card -> Long-term memory

Comment

Inspiration

I don't know if I simply get older and more senile (I am 31), but I keep forgetting facts important to me and my discussions or concepts I want to use in my thinking. To retain these I could use Anki flashcards, but the UI is old and the card creation process too manual in the age of AI. Thanks to Gemini I can finally atomate this process and make it way faster. I hope I'll never forget what the Sapir–Whorf hypothesis is!

The name conveys the intention: interconnecting minds - bridging the speed of AI and the universality of the human mind.

What it does

Intercortex is a cross-platform learning app that turns a word or concept into a focused flashcard. It uses an intuitive interface to make learning fun. Its FSRS spaced repetition for scheduled review is science-backed and used by leaders (Anki)

It supports:

Quick AI card generation from a single prompt A semantic follow-up flow when prompts are ambiguous Manual card creation and batch AI generation A visual, interactive home experience using draggable 3D bubbles Learning analytics (cards created, reviewed, streaks, success rate, token usage) Server-driven news/alerts and an activity log with retryable failures Privacy endpoints for user data export/delete

Why Gemini 3?

Gemini 3 Flash is the perfect model for my requirements: Knowledge about almost anything in the world, accessible in no time. No other model can be so quick to generate wildly different learning flashcards.

How I built it

With modern AI tools of course! Software development has changed significantly. While the app still has bugs this would have taken months, but instead implementation effectively took me 1-2 weeks. I was thinking about the concept for 2 years already though!

Technical info

Frontend:

Expo + React Native + Expo Router Three.js (expo-three / expo-gl) + cannon-es for the interactive bubble home UI AsyncStorage for local-first card storage ts-fsrs for review scheduling React Query for async data and mutation workflows Backend:

Node.js + Hono + Prisma + PostgreSQL AI routes powered by Gemini 3 Flash Preview Per-device quota/rate limiting, token/cost tracking, and request logging Admin APIs/UI for runtime AI settings, pricing, news management, and monitoring Deterministic simulation endpoints for testing AI error states Deployment:

Coolify + Docker on Hetzner Cloudflare for DNS/routing Split domains for app, API, and admin surfaces

Challenges I ran into

Ambiguous prompts Users type short words like “cell,” which can mean different things. Instead of guessing, the app returns a structured clarification question with options and then continues generation through a follow-up endpoint.

Cost and abuse control I added preflight and final quota checks, user/global rate limits, and token + micro-euro cost tracking per request so usage remains transparent and bounded.

Accomplishments that I'm proud of

Making my dreams a reality and providing value to the world. Full-Stack development and deployment of an app with an innovative UI.

What I learned

That the most recent AI software dev models are more than capable to surpass the speed of normal coder. Any app is just a few days/weeks of work away now!

What's next for Intercortex

  • Google Play/App Store submissions
  • Add more functionality like imports from anki or auto-summaries from wikipedia links
  • add more themes
  • Continue tuning Gemini prompts/settings from admin tooling without redeploys
  • Polish the gamified home experience and deepen progression mechanics
  • add ads and/or IAP for cash flow to cover AI and server costs
  • add automated tagging and categorization
  • add automated

Intercortex is my attempt to make AI-generated learning actually durable: fast enough for daily curiosity, structured enough for long-term memory.

I hope you'll like it!

PS: I will release the app version at submission time under https://hackathon-app.intercortex.app/ but the most recent version will be available at https://app.intercortex.app/

Built With

Share this project:

Updates