Kaleidoscope

What it does Instead of spending many weeks/months and hundreds of thousands of dollars testing different ads and messaging on real customers, use Kaleidoscope’s simulated customers instead. By comparing your existing customer tracking data (from Google Analytics, Shopify, Stripe) and your historical customer interactions (social media posts, TV ads, news articles, customer support calls, product reviews), my product learns how different segments of customers have historically responded to specific messaging and content. Given your business goals, it then creates swaths of AI agents and conducts controlled trials with them to find the most effective way for you to message your audience. Even household name brands find themselves short on marketing research spend and make disastrous PR blunders. Kaleidoscope catches them before the wrong person notices.

Inspiration

I was inspired by a friend who was working on his own startup, and he would always complain to me about how difficult it was to test public interest in his project. I figured that many companies would also be in the same situation. To me, this seemed like the perfect place to try and use AI to fix this issue.

How we built it

Built with NextJS, Prisma as ORM, AWS RDS as database, using fine tuned OpenAI as our AI model, and traditional machine learning techniques for causal inference and AB testing.

Challenges we ran into

As someone not very experienced with hackathons and also working alone on this project, I struggled a lot with getting familiar with some new technology. Connecting all the components of the project together also took a fair amount of time. I also obviously didn't have real data from marketing campaigns to fill the databases with, so creating sample entries for those for the purposes of demonstration was a little difficult.

Accomplishments we're proud of

I'm mostly proud to have been able to get a finished demo of the project that I worked on. A specific part of the project I was proud of was the recommendation section that suggested improvements to marketing campaigns.

What we learned

I learned a lot about what it takes to create an actual product in a short amount of time under pressure and how to best divide up my time in situations like this. I obviously also became much more familiar with all the technologies involved.

What's next

I want to reach out to various companies and see if they would be interested in a product like this. If so, I believe that it could be much more fleshed out with time and more people working on this project.