Be part of our every day and weekly newsletters for the most recent updates and unique content material on industry-leading AI protection. Be taught Extra
Amazon Net Providers (AWS) and Riot Video games revealed at its AWS re:Invent convention the winners of this 12 months’s Valorant Champions Tour Hackathon. This 12 months, the hackathon was the Esports Supervisor Problem, the place members use AWS’s generative AI and Riot’s intensive suite of esports knowledge to construct instruments for esports managers to scout expertise and construct new methods. Based on AWS, 3,200 folks participated within the problem, with the successful groups splitting a prize pool price $61,000.
The winners have been Adrian Tan and Christina Chen, with their VCT Staff Builder software. This instrument makes use of a digital assistant with a chat interface that gives, amongst different issues, customizable staff strategies primarily based on participant attributes and stats. The instrument makes use of LLMs and machine studying by way of Amazon Bedrock to create its strategies primarily based on accessible knowledge about gamers and groups, and contains “agent-specific team aggression ratings, player performance and map positional analytics.”
Ashwin Raghuraman, AWS senior options architect and one of many 12 judges of the hackathon, spoke to GamesBeat in an interview concerning the successful instrument: “Basically they were building an application which is hosted on AWS and the large language models are using Bedrock to train so that as a prospective team manager, user of the application, you could basically just have a chat interface where you can interact with the model and ask these prompts.”
Raghuraman stated it wasn’t simply the interface or using generative AI that gave the successful submission the sting: “But then also we wanted them to integrate some follow up prompts and create justification. One of the things we were judging is understanding the justifications for how the model is pulling the data. That is actually a big part of the hackathon, is getting the data in a format that it is easily accessible for both retrieval by the model, but also understandable by the human who is reading that retrieval of the data.”
Based on AWS, VCT Staff Builder gained the hackathon for “its clean UI and workflow, detailed analyses, statistics, breakdowns of players, deep dive into agent capabilities based on maps and recommended strategies, and fast response generation.” Because the creators say on the VCT Staff Builder Devpost web page: “We are proud of the way our team worked together to brainstorm ideas, solutions and workarounds for the challenges we faced. We believe we stretched the goals to develop a high performing application that is designed with the end-user’s needs in mind.”
Raghuraman added that Tan and Chin made a number of selections that gave their submission an edge over different candidates: “They made a lot of very smart decisions on how to use different large language models for different purposes… That was, like, one of the main things that stood out for the AWS side, and on the Riot side, the data visualization was really good. They were able to bring in their game knowledge, but also did a really good job of using all of the different data sources that we provided.”
Doug Stevenson, Riot’s director of software program engineering for esports tech, supplied some recommendation for future hackathon members in an announcement: “When participants invest time to appreciate and support the nuances of players everywhere, it clearly shows in their work. Not only do these submissions win, but the key learnings and takeaways help Riot make sure players win, too.”
