Quiz Arena AI

1
2
3
4
5
6
7
8
9
10
11
12

Inspiration

We wanted to solve a common problem in education: quizzes often tell students whether they are right or wrong, but they don't always help them understand why.

At the same time, creating good quizzes and reviewing student performance takes a lot of time for educators. So we asked: What if AI agents could support both teachers and students while keeping humans in control?

That idea led us to build Quiz Arena AI.

What it does

Quiz Arena AI is a real-time competitive quiz platform powered by five AI agents.

Quiz Generator creates quizzes from lecture materials.
Syllabus Agent scans course content and prepares quiz drafts while checking for duplicate topics.
Hint Master gives students helpful clues without revealing the answer.
Professor Q acts as a Socratic tutor after the game and helps students understand their mistakes.
Commentator Agent provides live game-show-style commentary based on what happens in the match.

Students can compete in real time, earn XP, maintain streaks, and progress through different ranks.

Most importantly, the AI does not replace the teacher. AI prepares and assists, while the educator makes the final decision.

How we built it

We built the frontend using React, TypeScript, Vite and Socket.IO, with a Node.js/Express gateway handling the real-time game.

For the AI layer, we used the Strands Agents SDK with Amazon Bedrock, and designed the system to run on Amazon Bedrock AgentCore.

The agents use different patterns such as tool calling, autonomous multi-step workflows, and conversational memory.

We also added AWS S3 for content storage, Firebase authentication for admin access, Docker for deployment, and GitHub Actions for automated testing and CI/CD.

Challenges we ran into

One of our biggest challenges was making multiple AI agents work reliably as part of one real-time application rather than as separate AI demos.

We also had to think about:

Keeping gameplay fast while AI features run in the background.
Preventing AI-generated quizzes from going directly to students without teacher approval.
Protecting answers from being exposed during gameplay.
Handling AI failures and fallback scenarios.
Making the interface accessible for different types of learners.

Accomplishments we're proud of

We're proud that Quiz Arena AI is more than a traditional quiz application with an AI chatbot added to it.

We built a complete multi-agent educational workflow where each agent has a specific responsibility.

We're especially proud of the human-in-the-loop Syllabus Agent, the Socratic Professor Q tutor, real-time AI commentary, live hints, accessibility-focused answer design, and our automated testing pipeline with 36 tests.

Most importantly, we designed the system around the Agents for Humans idea: AI handles repetitive work and provides assistance, while humans remain in control of important educational decisions.

What we learned

We learned that building an effective AI agent system is not just about choosing a powerful model.

The real challenge is designing the right role, tools, boundaries, and workflow for each agent.

We also learned a lot about Strands Agents SDK, Amazon Bedrock, AgentCore, real-time systems, AI safety, accessibility, and human-in-the-loop design.

What's next for Quiz Arena AI

Our next goal is to make Quiz Arena AI more personalized and useful for real educational environments.

We want to improve learning analytics, personalize questions based on student performance, expand the tutor's memory and learning guidance, and make the platform easier for educators to integrate into their existing courses.

Our vision is simple:

Don't use AI to replace the teacher. Use AI to give teachers more time to teach and students more opportunities to learn.