PartyRock_Hackathon - Toddler Learning Activities

PartyRock_Hackathon - Toddler Learning Activities

Entering into another hackathon using PartyRock!

Published Feb 13, 2024
Last Modified Feb 15, 2024
I have done a lot of hackathons in the past. I don't win super often, but they are a lot of fun and give me a chance to try something new with a bit of hope attached to them. This hackathon seemed like something completely different than what I have done before.
Most hackathons I have done were related to voice-based systems like alexa or bixby, but this was something different in that it's focus was AI. I haven't done a lot of AI based hackathons because for the most part I'm not sold that AI is the world-ending god-killing technology many people seem to think it is (at least not in it's current form). I have, however, played around with OpenAi, bedrock, and a ton of other AI based tools in the past, so I know a bit about prompt engineering.
What I really loved about this tool was how easy it was to use. Getting a new idea up and running was super simple, and giving access to a lot of different models from bedrock meant there was a lot more control about how fast and accurate responses needed to be. Some models are better suited for specific tasks, and some are better for open ended things, and this gave us both.
What I ended up creating was a little tool I thought would be cool called Toddler Learning Activities.
A screenshot of it
Toddler Learning Activities
I love doing things with my own toddler and going on mini-adventures, so this was near and dear to my own heart because it's the kind of tool I might use myself when trying to think up new things to do with her.
In the past I have done a lot of prompt engineering for RPG based games for my Alexa skills, so this was something completely different. A lot of knowledge translated, but I also had to learn a lot of new things about it (especially for image generation since Alexa is mostly voice based).
It was also a lot of fun to create, and honestly a bit shocking at just how easy it was. I definitely plan on spending a lot more time with bedrock to see what I can create!

What would I have done without this?

Honestly, I'm not sure I would build something quite like this with only the original tools that Bedrock offers. I love the bedrock API and it is extremely useful, but the biggest issue overall is cost.
A tool like this can be incredibly powerful and useful, and I can imagine tons of parents using it to get started in the right direction. But, when building it you essentially have to run an AI response, then use that to run another AI response, and then run many other AI responses. This is an incredibly powerful tool and use case, but that means every single interaction requires multiple API calls, which in turn each trigger a cost.
With Bedrock, I likely would have drastically simplified the design and relied more heavily on doing everything in a single API call and then parsing the solution into what I need to display. I still would have had to generate images separately, which means I wouldn't have generated nearly as many and they wouldn't have been as interesting or accurate.
I would also have had to think very early on about the cost of such a tool and how to monetize it. People are very used to not paying for things on the internet, and having to add some type of upfront cost to use it could be a dealbreaker. Adding advertising could help, but likely not enough to manage it fully. All in all, there are a number of shortcomings with trying to build this on my own that might have made it much more difficult and taken a lot more time to write custom code to do it.
What I would have done:
1) Build a custom UI with react to display the results.
2) Build an endpoint to take in user input and run the results through two API calls
3) The first API would have been to Bedrock Claude to get back some ideas in a response.
4) The second API would have been an image to use.
5) Display the image along with its represented suggestion and then have more suggestions down below.
It might have ended up being a slightly faster product, but it would have taken a lot of time to build and likely wouldn't have been worth it once you add the cost of doing it.

Community Impact

A lot of people have children and would want to search for a way to spend time productively with their children. Sure, they won't all grow up to be Einstein, but that isn't the point. A tool like this can give suggestions about things you can do with your child that are extremely low cost and easy to do.
I could see a lot of people using it to generate suggestions and spend time with their toddlers. I could also see it being reprised (or remixed) to create other suggestions for older children or adults (maybe ideas for a date night or something else).
Want to give it a try?
Here is my DevPost entry: