Inspiration
What it does
How we built it
Fully with Kiro. Giving requirements at first and then discussing with Kiro. Initially I wanted to use small AI vision libraries such as mediapipe or Yolo. But the results were underwhelming. Since I wanted to develop it fully with Kiro and not plays with training, wieghts... I decided to use Bedrock and Claude or Nova Lite2 to be more frugal. Another Option I might try is a fully local SmolVml from HuggingFace
Challenges we ran into
Early AI libraries were underwhelming. Kiro's suggestion were not always the best and being AI savvy helped a lot to guide it toward the right solution. A lot of back and forth to get a lively companion which look matched the right mood. After the initial requirements and all the kiro setups and development of tasks, I was unusure what to do so I switched back to vibe coding to tailor the app to my need.
Accomplishments that we're proud of
This is working well. It is a cute and fun app, with great and clean code, all without looking at the code myself. It developed in a short timeframe
What we learned
Initial requirements are important. But then we need to switch to a bit of vibe code to tailor the app. Need more experience on how to move further with an app and support it. It is an awesome tool but one needs to be careful how to use it on the long run and in a team.
What's next for TamagotchAI
More interactions and why not a version at the edge on a small pi device
Built With
- ai
- amazon-web-services
- bedrock
- claude
- kiro
- node.js
- nova
- python
- sdk
- smolvlm
- yolo
Log in or sign up for Devpost to join the conversation.