FlashLearn AI

An AI tutor that knows what you need to master next. Personalized learning paths that adapt, track progress, and transform beginners into experts. Real mastery. Real results.

Comment

Inspiration

I kept learning new tech, new skills yet I always felt lost about what I truly needed to master next. So we built FlashLearn AI: to finally bring clarity, direction, and real progress to anyone who's tired of guessing their way forward.

What it does

FlashLearn AI acts as your personal mentor. Set a skill goal and the platform evaluates your current knowledge, builds a personalized improvement plan, and continuously tracks your learning journey. It identifies your knowledge gaps and adapts the content in real time to keep you moving forward. FlashLearn AI primarily leverages an intelligent system of flashcards and quizzes to help you make steady progress and reveal blind spots in your expertise.

FlashLearn AI isn't designed to replace traditional learning platforms, it's designed to complement them.

The vision:

  1. Take a course on any platform (Udemy, Coursera, YouTube, read a book, etc.)
  2. Use FlashLearn to assess what you actually retained
  3. Identify your specific knowledge gaps (what you didn't master or poorly understood)
  4. Get a personalized remediation path focusing exactly where YOU need work
  5. Track progress on closing those gaps with adaptive practice

The key insight: After learning structured content, everyone has different weak points. FlashLearn quickly detects those gaps and lets you know exactly where to focus your efforts to reach mastery.

It's like having a personal tutor who:

  • Tests your understanding
  • Identifies what you missed
  • Creates custom practice focused on YOUR gaps
  • Adapts difficulty as you improve
  • Guides you on a personalized path toward expertise

How we built it

Architecture & Tech Stack

Frontend (React + TypeScript on Cloud Run):

  • React 18 with TypeScript for type-safe development
  • Vite for lightning-fast builds (< 3 seconds)
  • Tailwind CSS for rapid UI development
  • Redux Toolkit + Redux Persist for state management
  • Framer Motion for smooth animations
  • Nginx as reverse proxy for API routing
  • Deployed as containerized service on Cloud Run

Backend (Node.js + TypeScript on Cloud Run):

  • Node.js 22 with Express.js framework
  • TypeScript for compile-time type safety
  • JWT-based authentication with Firebase Admin SDK
  • RESTful API architecture with comprehensive error handling
  • Winston logging integrated with Google Cloud Logging
  • Deployed as containerized service on Cloud Run

Google Cloud Services:

  • Cloud Run: Serverless container platform hosting both frontend and backend
  • Firebase Firestore: NoSQL database for user data, learning objectives, and progress tracking
  • Firebase Authentication: User management and identity verification
  • Gemini AI API: AI-powered content generation for flashcards, quizzes, and assessments
  • Secret Manager: Secure storage for API keys and credentials
  • Cloud Logging: Centralized logging and monitoring

Adaptive Learning Engine:

  • Multi-factor difficulty adjustment algorithm (score, response time, consistency)
  • Spaced repetition system for optimal review scheduling
  • Real-time performance tracking and analytics
  • Knowledge gap detection with targeted remediation paths
  • Mastery level calculation (0-100 scale)

CI/CD Pipeline:

  • GitHub Actions for automated testing and deployment
  • Docker multi-stage builds (optimized images: 170MB backend, 50MB frontend)
  • Automated tests with Jest (backend) and Vitest (frontend)
  • Sequential deployment to prevent conflicts
  • Health checks post-deployment

Development Process

Week 1: Foundation

  • Architecture planning with Google AI Studio for brainstorming
  • Backend setup with Firebase integration and JWT auth
  • Gemini AI integration and adaptive learning algorithms

Week 2: Features & Polish

  • Frontend development with React and study interfaces
  • Advanced features (spaced repetition, analytics, validation quizzes)
  • Dockerization and Cloud Run deployment setup
  • Final testing, bug fixes, and submission

Key Technical Decisions

  1. Cloud Run as the Foundation: True serverless with container flexibility - scales to zero when idle, automatic HTTPS, built-in load balancing

  2. Firebase for Speed: Instead of setting up PostgreSQL and building auth from scratch, Firebase gave us production-ready database and authentication in hours

  3. Gemini AI for Content: AI-generated flashcards and quizzes mean infinite personalized content without manual creation

  4. TypeScript Everywhere: made refactoring safe and fast

  5. Multi-Stage Docker Builds

Challenges we ran into

1. Gemini AI Response Inconsistency

Challenge: Gemini API returned perfectly formatted JSON only 80% of the time. The other 20% included markdown code blocks, extra text, or creative variations.

Solution: Built a robust multi-strategy parser:

  • Try direct JSON parse first
  • Extract from markdown code blocks
  • Use regex to find JSON-like structures
  • Line-by-line parsing as fallback
  • TypeScript type validation for all outputs
  • Comprehensive error logging for debugging

Impact: Turned an unreliable integration into a production-ready feature with proper error handling.

2. Adaptive Learning Algorithm Complexity

Challenge: Initial simple threshold-based system (score > 85% = harder, score < 60% = easier) didn't capture the nuances of learning patterns.

Solution: Developed multi-factor algorithm considering:

  • Average score across recent sessions
  • Response time patterns (fast correct answers vs slow)
  • Consistency score (standard deviation of performance)
  • Review history and session frequency
  • Time since last study session

This solution will be really implemented next.

3. Authentication Architecture Decision

Challenge: Choose between Firebase client SDK (easier) vs backend-only JWT (more control).

Decision: Chose backend-only JWT with Firebase Admin SDK for:

  • Greater flexibility for future integrations
  • Better security control
  • Easier migration path if needed
  • Cleaner separation of concerns

4. Time Constraints While Working Full-Time

Challenge: Building a production-ready app in 2 weeks with only nights and weekends

Solutions Applied:

  • Ruthless feature prioritization (cut 60% of initial feature list)
  • AI-assisted development for 3-4x productivity boost
  • Focused weekends on deep work, weekdays on small tasks

Accomplishments that we're proud of

Technical Achievements

*Built and Deployed *

  • Complete full-stack application from concept to production in less than 2 weeks
  • Working nights and weekends while maintaining a full-time job
  • 8,000+ lines of production-quality TypeScript code

True Serverless Architecture

  • Both frontend and backend running on Cloud Run
  • Scales from zero to thousands of users automatically
  • $0 cost during development (scales to zero when idle)

Production-Ready AI Integration

  • Gemini AI generating high-quality flashcards, quizzes, and assessments
  • Robust parsing system handling AI response variations
  • Adaptive content generation based on user performance
  • Real-time difficulty adjustment using intelligent algorithms

Excellent Performance

  • < 2 second cold starts on Cloud Run
  • < 200ms average response time for most operations
  • Zero critical bugs in production since launch

Security Done Right

  • JWT-based authentication with Firebase Admin SDK
  • Google Secret Manager for all credentials
  • HTTPS everywhere (Cloud Run default)
  • Rate limiting and security headers (Helmet)
  • Encrypted data at rest (Firestore default)

Intelligent Adaptive Learning

  • Multi-factor difficulty adjustment algorithm
  • Spaced repetition for optimal review timing
  • Real-time knowledge gap detection
  • Personalized learning paths for each user
  • Progress tracking with detailed analytics

Personal Achievements

Introduced to Google Cloud Ecosystem

  • Gained production experience with Cloud Run
  • Learned Secret Manager, Cloud Logging, Container Registry
  • Understanding of serverless deployment patterns
  • Confident deploying complex applications to GCP

AI Development Expertise

  • Production experience with Gemini AI API
  • Prompt engineering for consistent outputs
  • Building robust parsers for LLM responses
  • Understanding AI limitations and mitigation strategies

What we learned

Technical Learnings

1. Cloud Run is Production-Ready for Real Applications

  • Not just for simple APIs, can handle complex full-stack apps
  • True scale-to-zero saves significant costs
  • Container-first approach provides ultimate flexibility
  • Built-in features (HTTPS, logging, monitoring) accelerate development
  • Cold starts (< 2s) are acceptable for most applications

2. Managed Services Accelerate MVP Development

  • Firebase setup: 1 hour vs 15+ hours for custom auth + database
  • Gemini API: 2 hours to integrate vs 40+ hours creating content
  • Secret Manager: Secure by default vs complex custom solutions
  • Trade control for speed, perfect for hackathons and MVPs

Process Learnings

3. AI-Assisted Development is Transformational

  • 3-4x productivity boost when used correctly
  • AI writes boilerplate, human makes decisions
  • Review everything, AI isn't perfect
  • Combined with human expertise = powerful multiplier
  • Essential tool for time-constrained projects

4. Modern Tools Have Lowered the Barrier Dramatically

  • Solo developer can build what required a team 5 years ago
  • What would've taken 3 months took 2 weeks
  • What would've cost $50k+ costs $5/month
  • Technical barrier has collapsed, psychological barrier remains

Strategic Learnings

5. Validate Before Building

  • Used Google AI Studio for concept validation
  • Saved days of potential wasted development
  • Test core assumptions (Can Gemini generate quality content?)
  • Start coding only after validation

6. Google's Integrated Ecosystem is Powerful

  • Cloud Run + Firebase + Gemini work seamlessly together
  • No integration headaches between services
  • Consistent authentication and authorization patterns
  • Single bill, single console, unified logging

What's next for FlashLearn AI

Platform Integrations

  • Learning Platform Partnerships (Udacity, Coursera, Udemy, Pluralsight, edX)
    • Direct integration with course completion data
    • Automatic assessment after finishing courses
    • Identify specific knowledge gaps from completed material
    • Provide personalized next-step training recommendations
    • Certificate of true mastery (beyond just completion)
    • "You finished React course on Udacity → FlashLearn identifies you struggle with Hooks → Recommends targeted practice + related courses"
  • Calendar integration (Google Calendar, Outlook) for study scheduling
  • Zapier/Make integration for workflow automation

Improved Adaptive Algorithms

  • Machine learning model trained on user performance data
  • Better prediction of optimal review timing
  • Personalized difficulty curves based on learning style
  • Automatic topic recommendations based on progress

Mobile Applications

  • React Native apps for iOS and Android
  • Native features: push notifications for review reminders
  • Offline mode for studying without internet
  • Home screen widgets showing daily progress

Social Features

  • Study groups and collaborative learning paths
  • Share flashcard sets with friends or publicly
  • Leaderboards and challenges to gamify learning
  • Community-curated learning paths

Enhanced Content

  • Support for multiple AI models (Claude, GPT-4) for content variety
  • Image support in flashcards (diagrams, code with syntax highlighting)
  • Audio pronunciation for language learning
  • Video embedding for multimedia learning

Advanced Analytics

  • Detailed learning curve visualization
  • Identify knowledge gaps automatically across multiple topics
  • Predictive modeling (when will you master this topic?)
  • Export detailed reports (PDF, CSV)

Monetization Strategy

  • Free Tier: Current core features, limited to 100 flashcards/month
  • Pro Tier ($9.99/month): Unlimited flashcards, priority AI generation, advanced analytics
  • Team Plans ($49.99/month): For companies and educational institutions with admin dashboards

AI Tutoring & Mentorship

  • Chat with AI tutor about specific topics
  • Get explanations when stuck on concepts
  • Ask follow-up questions on flashcards
  • Personalized study plan recommendations

Course Marketplace

  • Curated courses by domain experts
  • Certification programs with verified assessments
  • Revenue sharing with content creators

Internationalization

  • Support for 20+ languages
  • Localized content generation
  • Multi-language flashcards for language learners
  • Regional AI models for better cultural context

Vision: The Future of Personalized Learning

FlashLearn AI represents a shift in how we approach skill mastery. Instead of one-size-fits-all courses or generic flashcard apps, we're building the intelligent layer that sits between learning resources and true expertise.

Our goal: Make personalized learning accessible to everyone, not just those who can afford personal tutors or expensive bootcamps.

Our mission: Help one million learners identify their knowledge gaps and create personalized paths to mastery.

Acknowledgments

  • Google Cloud team for organizing this hackathon
  • The Google AI team for Gemini API
  • Everyone who provided feedback and encouragement

Thank you for considering FlashLearn AI!

Making personalized learning accessible to everyone, one flashcard at a time. 🚀

Built With

Share this project:

Updates