VCT Hackathon Project

A digital assistant for the Valorant Champions Tour that utilizes the latest stats and info to provide users with detailed analysis of players and suggest team compositions.

Comment

Inspiration

  1. We are fans of the valorant pro scene, so if felt right for us to work on this project for VCT
  2. Our interest in machine learning and web-dev felt applicable for this project

What it does

Our app takes user prompts as well as filter parameters in order to make RAG calls to our model (claude-sonnet-3) and knowledge base using aws lambda.

How we built it

We built our frontend in React and utilized python and boto3 for our lambda function. We also utilized python for datascraping and compilation from these places: vlr.gg, riot games data, youtube, gamer.gg, and redbull.com. Our app is hosted on an EC2 instance.

Challenges we ran into

  1. Lack of text information on playstyles and team roles such as igls
  2. Issues with citation and reference generation from RAG calls
  3. Web app hosting and throttling

Accomplishments that we're proud of

  1. Datascraping and metadata tagging of vlr.gg/stats
  2. Working RAG calls with editable filters
  3. Clean and concise UI

What we learned

  1. Organizing and allocating tasks for different aspects of the project
  2. Communicating well between different teams

What's next for VCT Hackathon Project

  1. Get RAG citations working consistently and neatly displayed in the frontend after calls.
  2. Add databasing for better player retrieval, as well as persistent and flexible team creation
  3. Develop more advanced metrics for player and agent synergies
  4. Find more relevant textual data regarding team and player performances and styles to fine-tune our model

Built With

Share this project:

Updates