We're a think tank, so it only makes sense to put everyone in a room and have them brainstorm random ideas. We then build quick and dirty solutions to implement these ideas during the rest of the day or the weekend! Some of these have also turned into competitions and other larger projects!
A few of the brainstorming ideas turned into internal competitions because they didn't need more than 1-2 people to build/implement them. So, we had everyone make their own entries and found the best!
AI Connect Four Bot
Some of our team members love AI that play games. We decided to take that opportunity to build AI bots that play a simple game
"Connect Four", and then pit them against each other!
Naru hosted the competition by building a standardized framework and providing everyone else an SDK/template to work off of.
NOTE: John tried to cheat by building a bot that attempts to hack the competition framework, but Naru's smart - he built a security system to prevent that. PS. John didn't win.
Some of the AI bots used artificial neural networks to learn strategies, others used evolutionary algorithms, and some - very simple pre-programmed strategies.
Andrew won this competition, with
Marcel only two wins behind.
AI Server Supervisor
Ok, so this competition came into existence because we wanted to encourage our team to develop novel AI programs for Stratum.
Stratum is a project we're working on that uses artificial intelligence and predictive scheduling to optimize how containers are allocated on servers.
In this competition, Naru (guess he's always the host. That prevents him from ever losing.) built a job request and cloud worker simulator, along with a supervisor SDK/template for the team members. Everyone needed to build their own AI and the goal was to minimize the total server cost over various task profiles.
For simulations running over 1 day, 1 week, and 1 month,
Derek (our newest team member at the time!) got the lowest total server cost and cost per job - a whole 8.8% lower than the nearest competitor!
For the 6 month simulation,
Andrew's AI resulted in the lowest server cost and cost per job - 8% better than the nearest (Derek's). It was really interesting how Andrew's solution was able to perform better in the long run.
Clay and
John get honorable mentions for the performance of their bots - not too far behind the winners.
We learned so much from this and incorporated components from each of the submissions into Stratum!