Inspiration

My daughter said to me "no one trusts stuff made with AI" and I thought, what if I build a way to generate lots of AI content but with a really easy way for a human to check it for quality and correctness, before it's used.

What it does

It allows you to generate lots of AI content for a given purpose (e.g. for games) and allows you to approve or reject the content, by showing you how it will be rendered when used. I've also built an example consuming application - https://newgameplease.com/ - which uses the approved AI content to render different types of mini-games.

How we built it

CDK! 10 Lambdas, API Gateway, Stepfunctions, React app on S3 & Cloudfront, Dynamodb, Cognito

Challenges we ran into

  • I hadn't used CDK before on a personal project, so that was a tiny bit of learning-curve, but once I got up and running it was super useful!
  • Getting GenAI to behave! I use Bedrock (Claude model) to generate content, so there was a bit of prompt engineering required - which to be honest was more me not defining in enough detail what I wanted!

Accomplishments that we're proud of

I only found about the hackathon on the 14th June, so I'm proud of getting so much done in 2 weeks! A lot of late nights, but it was very rewarding to see it all come together.

The solution is fully deployed and working!

What we learned

Just how easy it is to build complete secure solutions using CDK, Lambda and other AWS services.

What's next for Human In The Loop AI Content Generator

To make my sample games better!

Share this project:

Updates