Valorant Manager - Ben Tenz

This project uses AWS resources such as Bedrock, Lambda, S3 and more to create an LLM that can generate and explain successful teams based on player statistics.

Comment

Inspiration

We loved the idea of creating a project that touches on something all members of the group enjoyed playing; Valorant, more notably, its competitive scene. As a group, we decided that it would be cool to learn AWS Bedrock and implement the skills we had before this hackathon to create the best possible LLM that could create cool teams that we, as well as many others in the player base, would imagine as the dream team.

What it does

Our project leverages the services AWS provides by using Amazon Bedrock as an agent builder by having it interact with different services such as S3 and Lambda to provide key information our agent needs to create the best team based on given criteria. It uses the information scraped from VLR, stored in our S3 bucket, to make predictions on what makes a player with the Lambda function providing an efficient method of filtering out any bad players. It then makes its own decision on what kind of team to make by prompt engineering it to identify key components of a team such as IGL, and player roles in the team.

How we built it

We used Amazon Bedrock as our foundation to create our agent and its workflow. Additionally, within the agent, we created a lambda function used to effectively filter out player data without the LLM having to do it. Using an S3 bucket we can store the web-scraped data retrieved using Beautiful Soup. We are also hosting the webserver using Amazon Lightsail and Node.js.

Challenges we ran into

Many of the obstacles we came across involved learning AWS, especially when working with IAM roles and permissions or just figuring out how to navigate around it in general. Additionally, there were some issues with data in VLR and some information from certain regions did not exist occasionally there would be edge-case situations of teams forfeiting matches which would mess up our data.

Accomplishments that we're proud of

We are really proud of our front end as well as our ability to learn AWS features through project-based learning. We are happy that our product feels complete and the efforts we put into making this resulted in a working project.

What we learned

The main thing we learned was how to utilize AWS services and integrate them into our project.

What's next for Valorant Manager

Eventually, we would love to make the project automatically scrape data as new information comes out instead of manually web-scraping data every time we create a new iteration

What would we do if we didn’t have throttling issues

  1. Reducing the tokens used to reduce the probability of more throttling issues, this can be done by optimizing the size of the prompts used and identifying whether they work as intended.

  2. Work around having follow-up prompts. Currently, our front-end is meant to create teams, and not to have follow-up prompts. With some troubleshooting with our Agent, we can update our front-end to identify which action group is running and adjust accordingly.

  3. Troubleshooting the new prompts provided. We originally wanted to find out what our current model can and can't do and make the adjustments necessary to meet the requirements for all the prompts mentioned.

  4. Optimizations. Currently, our agentic workflow uses two agent calls, but this can be shortened to one as we recently learned how to access the agent's trace, making it more efficient in gathering agent information for both our front-end and back-end.

Built With

Share this project:

Updates