Radiant Forge

Inspiration

The inspiration for Radiant Forge came from the growing complexity of team composition in competitive Valorant.

We wanted to create a data-driven tool to help teams, coaches, and players optimize their performance by scouting potential players, analyzing their strengths, and building balanced rosters.

With AI tools and player data becoming increasingly accessible, Radiant Forge was designed to bridge the gap between raw stats and meaningful, actionable insights.

For the name Radiant Forge, the top players in Valorant achieve the rank of Radiant, and Forges are where unique combinations of metals get bonded together to build amazing things, so we took both of those and made Radiant Forge.

We also really like the Radivores, that is also why we included Wingman in the app, let us know if you can spot the amazing Wingman (there are 2 places where Wingman is)

What it does

Radiant Forge is a Valorant team composition chatbot that is pwoered by AWS Bedrock.

Radiant Forge's abilities are enhanced by the scraped player statistics from VLG.gg and we also leveraged Crew AI to synthesize this data into comprehensive player profiles.

These profiles are then used in a Retrieval-Augmented Generation (RAG) system to optimize team compositions while following the user's requirement, ensuring that the best possible combination of players is selected based on their performance and compatibility.

How we built it

The project was built using several technologies:

• Data scraping: Player data was scraped from VLG.gg, focusing on tournament performance.
• Crew AI & Gemini 1.5: Synthesized raw player stats into readable profiles by developing Player Scout, Reporter, and Manager agents.
• AWS Bedrock: Utilized to manage knowledge bases for different levels of players (Pro, Challengers, Game Changers) and the powerful agents to deliver team compositions.
• Streamlit: Built the web interface to display the player profiles and team compositions.
• AWS Lambda & S3: Used for processing queries and storing player profile data.

Challenges we ran into

• One of the major challenges was identifying the most relevant player statistics for optimal team composition.
• Scraping and categorizing data from VLG.gg was another hurdle, especially when filtering players by combat score and tournament participation.
• Additionally, AWS Bedrock's limitation on the number of knowledge bases attached to an agent required us to implement a lambda solution for querying multiple knowledge bases.

Accomplishments that we're proud of

We successfully built an AI-driven system that turns raw data into meaningful player profiles and optimizes team compositions using a RAG-based approach.

The web application presents this data in a streamlined UI, making it easy to interact with.

We also overcame infrastructure challenges and built agents that effectively synthesize complex Valorant stats into structured profiles.

What we learned

We learned a lot about the intricacies of scraping and processing competitive esports data, as well as the potential of RAG systems in building AI-driven solutions for team optimization.

Our work with Crew AI and Gemini 1.5 also deepened our understanding of AI synthesis for player profiling.

Lastly, we gained experience in utilizing AWS Bedrock and handling infrastructure constraints with Lambda and S3.

What's next for Radiant Forge

In the future, we plan to scale Radiant Forge by expanding the dataset to include more players and refining our RAG approach.

We also want to implement advanced filtering features that will allow users to build custom teams based on specific performance metrics.

Additionally, experimenting with more sophisticated AI models to improve profile retrieval and synthesis is a key goal.

Built With

• amazon-web-services
• aws-bedrock
• aws-lambda
• crew-ai
• gemini
• python
• streamlit