Vedira

A helpful, always-available app to help you learn anything you want!

Comment

Inspiration

As all four of us were students, we often used generative AI tools to help with our studies, whether it was brainstorming ideas, writing papers, or solving problems. These AI tools proved to be immensely helpful in making our academic work more efficient and personalized. However, as we continued using these tools, we realized that the true potential of AI lies not just in assisting with tasks but in offering tailored, adaptive learning experiences.

This realization inspired us to create an app that could harness the power of generative AI to provide personalized, on-demand education. We envisioned a platform that adapts to individual learning needs, offering customized lessons, quizzes, and flashcards based on each learner's pace and prior knowledge. We knew that such a solution could greatly benefit students and learners around the world.

With the AWS Lambda hackathon on the horizon, we knew this was the perfect opportunity to bring our vision to life. Using AWS Lambda as the heart of our serverless architecture, we could scale the app to deliver personalized learning experiences to anyone, anywhere, in real-time. Our own experiences as students using AI tools motivated us to build something that could empower others to learn more effectively, no matter their background or learning style.

What it does

Vedira is a mobile-first, AI-powered learning platform that customizes the educational experience for each user. With Vedira, users can quickly generate personalized course structures, quizzes, flashcards, and more, simply by entering what they want to learn, how long they want to spend, and any prior knowledge, and within a minute, they'll have a customized course plan! The app generates:

  • Personalized course structures with chapters and lessons tailored to your learning goals
  • Interactive lesson content with markdown formatting, code blocks, and rich media
  • Multiple-choice quizzes to test your understanding after each lesson
  • Flashcards for spaced repetition and memory reinforcement
  • Real-time progress tracking so you can see your learning journey unfold
  • Course generation from files - upload PDFs, documents, or paste text to create courses from existing content

How we built it

Frontend: Flutter Mobile App

  • Cross-platform development with Flutter for iOS, Android, Web, and Desktop
  • 10 comprehensive screens covering authentication, course management, and interactive learning
  • 11 specialized services handling everything from API communication to offline data management
  • 9 data models for courses, lessons, quizzes, flashcards, and progress tracking
  • 12 custom widgets and components for consistent, beautiful UI across the platform
  • Provider pattern for clean state management and reactive UI updates
  • Clean design with dark/light mode support
  • Secure authentication with JWT tokens and automatic refresh handling
  • Rich content rendering with markdown support, syntax highlighting, and interactive elements

Backend: AWS Serverless Architecture

  • 22 specialized Lambda functions handling everything from authentication to content generation
  • AWS Step Functions orchestrating complex workflows for parallel content generation (lessons, MCQs, and flashcards)
  • DynamoDB (2 tables) for scalable course and user data storage with GSIs for efficient querying
  • S3 buckets (3 buckets) for lesson content, quiz questions, and AI-generated course images
  • AWS CDK for infrastructure as code, making deployment and scaling seamless
  • AWS Cognito for secure user authentication and authorization
  • API Gateway with CORS, rate limiting, and request/response transformation

AI Integration

  • Multi-provider LLM support with Google AI Studio (Gemini) and AWS Bedrock (Claude) integration
    • We used the Bedrock Access Gateway to connect to Bedrock, allowing us to utilize the OpenAI API while also having access to Bedrock, Google, and OpenAI models (although we ultimately did not use the OpenAI models!).
  • Intelligent fallback mechanisms ensuring content generation continues even if one AI provider fails
  • Rate limiting and intelligent staggering to handle API quotas gracefully
  • Document processing with text extraction from various file formats
  • Structured output validation ensuring consistent, high-quality educational content

Architecture Highlights

  • Event-driven processing where course generation triggers parallel workflows
  • Microservices approach with each Lambda function serving a single purpose
  • Real-time status updates using polling for generation progress with optimized API calls
  • Parallel processing where MCQs and flashcards generate simultaneously after lesson completion

Challenges we ran into

Despite the excitement of building this app, we encountered several challenges along the way:

Complex Orchestration

Managing the parallel generation of lessons, MCQs, and flashcards while providing real-time progress updates required sophisticated Step Function workflows with proper error handling and retry mechanisms across several different Lambda functions.

Mobile-First AI Experience

Creating a smooth, responsive mobile experience for AI-generated content meant solving challenges around:

  • Long-running AI operations with progress indicators and user engagement
  • Efficient content caching and offline access across multiple content types
  • Responsive markdown rendering with code syntax highlighting and rich media
  • Smooth animations during content generation without blocking the UI
  • Cross-platform authentication that works on iOS, Android, Web, and Desktop

Sample Login Credentials

For testing purposes, you can use the following sample credentials:

Email: light1507041@gmail.com
Password: Test@123

Serverless State Management

Coordinating state across multiple Lambda functions and managing generation status in DynamoDB while ensuring consistency and real-time updates to the mobile app across different user sessions.

Content Quality and Consistency

Ensuring AI-generated educational content maintains quality, proper formatting, and educational value across different topics and difficulty levels while handling edge cases in markdown parsing and code block rendering.

Accomplishments that we're proud of

Complete Learning Ecosystem

Built a full-featured learning platform from ground up, including course creation, content consumption, assessment, and progress tracking - all powered by AI and running completely serverless.

Production-Ready Serverless Architecture

Implemented a sophisticated serverless backend with 22 Lambda functions, comprehensive error handling, parallel processing, and complete infrastructure as code with AWS CDK.

Beautiful Cross-Platform Mobile Experience

Created a polished, native-feeling mobile app with 10 screens, smooth animations, adaptive theming, and intuitive user experience that works seamlessly across iOS, Android, Web, and Desktop.

Real-Time Generation Tracking

Solved the complex problem of providing real-time feedback for long-running AI operations across multiple parallel workflows, keeping users engaged during the content generation process. All lessons in a chapter generate simultaneously, reducing wait times for users. Post-generation, lessons are immediately available to users while MCQs and flashcards wrap up generation in the background.

Robust AI Integration

Successfully integrated multiple AI providers with intelligent fallback, rate limiting, and error recovery, ensuring reliable content generation even under challenging API conditions.

Scalable Architecture

Built with AWS best practices that can automatically scale from a single user to thousands without infrastructure changes, paying only for actual usage.

What we learned

  • Serverless-first thinking: Using AWS Lambda and Step Functions allowed us to create an event-driven, scalable solution without worrying about managing infrastructure.
  • Mobile AI UX is different: Designing a smooth AI-powered experience on mobile requires careful consideration of user feedback loops, offline functionality, and engagement.
  • Rate limiting is critical: Managing API quotas and ensuring seamless operations was crucial for a scalable AI-driven app.
  • Content Quality and Iteration: Continuously improving AI prompts and validating outputs was key to ensuring high-quality educational content.

What's next for Vedira

  • Adaptive learning paths that adjust to user performance and conceptual pain points
  • Collaborative features for study groups and peer learning
  • Advanced analytics providing detailed insights into learning patterns and progress

VEDIRA represents the future of personalized, on-demand education powered by AI. With AWS Lambda at its core, we’re able to provide a scalable, efficient, and adaptable learning experience for learners everywhere.

Built With

Share this project:

Updates