-
-
SketchBox Home Page
-
Fetch.ai powered Game Master - will smartly route you to a game it picks for you
-
Collaborative Canvas - using websockets, all connected users will draw on the same canvas and AI will guess the drawing
-
Storytelling Game Mode - using websockets, all users can take turns sharing prompts to create AI generated scenarios and images
Inspiration
We have grown up playing online games, especially during the quarantine era. These games are classic and we believe they are great for passing the time with friends. We wanted to elevate that experience with AI, creating a seamless interaction on our platform as well as enhanced versions of the games we love.
What it does
Our application has three games: Competitive Skribbl.ai, Collaborative Skribbl.ai, and Bards. Competitive Skribbl.ai is a game where players try to compete with each other to get an AI to figure out their prompt, Collaborative Skribbl.ai is a game where people can collectively draw on a canvas to get an AI to recognize their prompt, and Bards is a multiplayer story generator that generates stories which anyone can contribute to.
How we built it
Using Fetch.ai as the brain of our infrastructure and core of our games, we route any player to a game that they enjoy and create bots composed of uAgents. We also leveraged Gemini’s expansive context window capacity to create rich multiplayer stories that never need to end, as well as to guess drawings the user drew on screen in real time. The underlying infrastructure was composed of web sockets to enable the online multiplayer aspect of our game. Additionally, we were able to publish a fine tuned model to the Intel LLM Playground using a dataset that would help increase our bot quality.
Challenges we ran into
Our original goal was to build our LLMs on Intel’s Developer Platform, however, we encountered technical issues with deployment. Our models were trained and fine-tuned on the platform, but we lacked sufficient compute power, and sadly had to pivot to Fireworks AI for image generation.
Accomplishments that we're proud of
We were able to implement websockets and fully functional games in a short amount of time along with having resolved several pesky bugs.
What we learned
We learned how to manage larger scale applications using web sockets. We also learned how to fine tune models to our liking so that the outputs are more desirable.
What's next for SketchBox
More integration of AI agents as well as a larger array of games so that the experience can be completely immersive and managed by an AI agent orchestrator.
Intel Developer Cloud Challenge Documentation: https://docs.google.com/presentation/d/1IoQqBpQZax6rQgZ-EIRzR9ySi4cDc8vgujpdLi4aHH8/edit?usp=sharing
Log in or sign up for Devpost to join the conversation.