CoLearn

Many quit online courses due to low motivation. We use Gemini 3 to generate courses and add social mechanics: join a class, chat with peers, and compete—turning solo learning into a team game.

Comment

Inspiration

Modern education often suffers from a lack of personalization—students are forced to follow rigid curricula that do not take into account their individual interests and learning pace. We were inspired by the idea of democratizing access to knowledge and creating a platform where everyone can learn what they are truly interested in, in a format that is convenient for them. Our goal is to make education as personalized as a private tutor, but accessible to everyone.

What it does

CoLearn is an intelligent educational platform that uses artificial intelligence to create personalized courses on any topic. Users can: 1) Generate unique courses using AI based on their interests. 2) Take structured lessons with rich content, examples, and exercises. 3) Communicate with an AI assistant for additional explanations and support 4) Track their progress and complete assignments with automatic checking 5) Learn in all available languages with gemini3

How we built it

We built CoLearn on a modern tech stack: 1)Frontend: React + TypeScript for a responsive and type-safe interface 2)AI Integration: Google Gemini API for course generation and an intelligent chat assistant 3)Styling: Modern CSS with gradients, animations, and responsive design 4) Localization: Built-in i18n system for supporting multiple languages 5)State Management: React hooks for efficient application state management 6)Development was iterative: first, we created the core course generation system, then added a testing system and chat functionality, and polished the UX/UI.

Challenges we ran into

1)AI generation quality: The first versions of the courses were too superficial. We had to carefully work out the prompts for Gemini to get detailed and structured content. 2)State management: Synchronizing user progress between different components required a careful architecture. 3) Parsing and formatting: Converting Markdown content from AI into a beautiful visual format required additional logic. 4)Mobile responsiveness: Adapting the interface for small screens without losing functionality was a challenging task.

Accomplishments that we're proud of

1) Fully functional AI course generation — users can create courses on literally any topic in seconds. 2) Smart testing system with automatic checking and feedback. 3) Beautiful and modern design that is pleasant to use 4) Multilingual out of the box — the platform is accessible to a wide audience. That is why we have made the interface available in two international languages (English and Russian) 5) End-to-end user experience — from course creation to progress tracking, everything works seamlessly

What we learned

1) Working with LLM requires careful prompt engineering to get quality results. 2) UX is super important in educational apps—even small details affect engagement. 3) TypeScript saves the day—typing helped avoid a bunch of bugs when working with dynamic data. 4) The importance of iterations — the first version is always far from ideal, and you need to be prepared to rework it. 5)Balance between automation and control — AI is a powerful tool, but you need mechanisms for customization.

What's next for CoLearn

1)Social features: ability to share courses with other users and create communities around topics Advanced analytics: detailed progress reports with recommendations for improvement 2)More content types: video generation (via veo3 integration), interactive images (by nanobanana) 3)Certification: issuance of course completion certificates 4)Integration with external sources: use of real articles, research, and videos to enrich courses 5)Voice assistant: ability to interact with AI through voice

Built With

Share this project:

Updates

posted an update —

What we added and fixed since the first version: 1)Moved to a real backend with Supabase We upgraded from a simpler setup to persistent cloud data with structured tables for profiles, courses, chat, gamification, classes, invites, and class messages. 2)Implemented full authentication and user profiles We now support proper sign up/sign in, session-based current user loading, profile updates, usernames (@handle), and avatar persistence. 3)Launched social learning with Class features Users can create classes, search and invite members, accept/reject invites, chat in class rooms, and view a class leaderboard. 4)Built competitive Arena modes We added AI Duel and Boss Fight with rounds, HP mechanics, difficulty levels, rating/tokens, and XP rewards. 5)Expanded gamification into a full progression system XP, levels, streaks, daily login rewards, and badge unlocks are now deeply integrated across lessons, modules, courses, duels, and boss fights. 6)Added certificates and PDF export Users can generate completion certificates and export lessons/courses to PDF for offline learning and sharing. 7)Strengthened AI reliability and response handling We added model fallback logic across multiple Gemini models and improved JSON parsing/validation for generated content. 9)Improved progress consistency and UX quality Course/module/lesson progress is recalculated and saved reliably, while mobile navigation and responsive behavior were significantly polished. 10)Improved security and production readiness We introduced a Supabase migration with Row Level Security policies to enforce safe, role-appropriate data access.

Log in or sign up for Devpost to join the conversation.