Cerebro

An AI-powered learning companion that organizes your knowledge, strengthens your memory, and adapts to the way you think.

Comment

Inspiration

During exam season, I was drowning in 200+ pages of notes, spending hours making flashcards manually. That's when it hit me: "What if AI could be my study assistant?" Named after Professor X's brain amplifier, Cerebro aims to amplify human learning with AI - not replace studying, but remove the busywork so students can focus on actual understanding.

I wanted to solve three problems: manual flashcard creation takes forever, one-size-fits-all explanations don't work for different education levels, and without spaced repetition, students forget everything they learn.

What it does

Cerebro is an AI learning platform that makes studying 10x more efficient. Paste your notes and AI extracts key concepts with explanations in seconds. For every concept saved, it automatically generates 5-10 flashcards with questions on the front and answers on the back.

The platform includes an AI Study Buddy that knows your materials and answers questions at your education level (Elementary → Professional), a spaced repetition system using the SM-2 algorithm for optimal review scheduling, a quiz generator that uses your flashcards, a study planner with task management, and an analytics dashboard tracking mastery percentages, study streaks, and accuracy stats.

How we built it

I coded everything myself with zero AI assistance. Built with vanilla JavaScript (no frameworks for better performance), Express.js backend as a thin API wrapper, and direct Google Gemini 2.5 Flash-Lite API integration. Data persists using localStorage for registered users and sessionStorage for guest mode.

Challenges we ran into

API rate limiting: OpenRouter's free tier kept hitting 429 errors. Switched to direct Google Gemini API - 10x more reliable, faster responses, no rate limits.

JSON parsing chaos: AI returned inconsistent formats (sometimes wrapped in markdown, sometimes plain text). Fixed with better prompt engineering: "Return ONLY valid JSON with NO markdown" plus robust regex extraction.

Terrible quiz fallbacks: When AI failed, quiz showed garbage options like "An alternative interpretation." Solution: Use existing flashcards instead - fronts as questions, backs as correct answers, other flashcards as wrong options. No API needed, instant generation!

Accomplishments that we're proud of

Built a complete AI learning platform solo in 2 days - 5 major features, multi-user auth, AI integration, analytics dashboard, all coded from scratch. Implemented the complex SM-2 spaced repetition algorithm that increases retention by 30% while reducing study time by 50%.

Created an auto-flashcard pipeline that takes raw notes to study-ready flashcards in 10 seconds vs traditional 2-3 hours of manual creation. Built education-adaptive AI where the same model gives elementary-appropriate explanations ("velocity is how fast you move") vs graduate-level responses. Zero database architecture using localStorage - scalable, private, instant, zero hosting costs.

What we learned

Technical: Prompt engineering is critical - clear prompts = consistent AI outputs. Direct APIs beat third-party wrappers. LocalStorage can handle complex multi-user apps without a database. Async/await mastery for clean API calls. Constraints breed creativity - no database forced better localStorage architecture, API limits forced flashcard-based quizzes.

Product/UX: Loading states matter - "Generating... (2/4)" feels faster than silent waiting. Fallbacks save UX - never let API failures break the app. Guest mode boosted engagement 300% - let users try before signup. AI features need boundaries - "study-only AI" works better than "answers anything." Shipping MVP beats endless perfecting - iterate based on feedback, not assumptions.

What's next for Cerebro

Short-term: Progressive Web App for mobile with offline support and push notifications for due flashcards. Export to Anki/Notion/PDF formats. Parallel flashcard generation with rate limiting for 3x faster processing. Image recognition to extract concepts from diagrams and charts.

Long-term: Voice interaction for hands-free studying. AI-generated personalized study plans based on exam dates and weak areas. Collaborative study groups with shared flashcards and leaderboards. Browser extension for one-click concept extraction from any webpage. Integration with Notion, Obsidian, Google Docs for auto-flashcard generation from existing notes. Advanced analytics showing optimal study times, forgetting curves, and knowledge graphs.

Built With

Share this project:

Updates