VCT Team Composer

• 

  Architecture

Inspiration

I wanted to learn more about how I can utilize LLMs in complex tasks

What it does

The app uses game data to compose a new Valorant team according to the user prompt and assigns roles to each team member based on the prior stats.

How we built it

I used Amazon Bedrock, LlamaIndex, Polars, and Gradio to build this application. I also used JupyterLab and Python scripts to clean and transform the data so the tools could utilize it easily and efficiently.

Challenges we ran into

Data preparation was the main challenge since I had to come up with the attributes that are informative for the LLM model but balance the amount of data I sent to it to keep the cost low. This involved a lot of testing and re-building the dataset.

Accomplishments that we're proud of

I'm proud of what I learned throughout the project and the learning paid off in the end!

What we learned

Reasoning agents, using AI tools in agentic workflows, using the Amazon Bedrock platform.

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

Like most other participants I have faced the "GenAI Throttling Problem" which made my application unusable at the finalists' stage. I have included the additional update document on the potential modifications to accommodate the finalists' prompts.

Built With

• amazon
• bedrock
• delta
• gradio
• llamaindex
• mistral
• polars
• python