Vanguard's Worst Nightmare

Inspiration

As long-time Valorant players and dedicated Esports enthusiasts, we've been immersed in competitive gaming for over a decade! Our passion for the VCT and its vibrant community inspired us to merge our technical skills with our love for the game, crafting a product that benefits Valorant coaches, managers, and fans alike.

What it does

Our LLM-powered digital assistant delivers data-driven team recommendations for the VCT International, Challengers, and Game Changers leagues. Whether building all-star teams or discovering rising talent, the assistant provides insightful guidance to assemble the perfect squad.

How we built it

We utilized Amazon Bedrock for LLM integration, CDK to orchestrate our infrastructure, and Streamlit to create an intuitive user interface. At its core, our assistant leverages Claude Sonnet as the foundational model, enabling powerful, context-aware responses.

Challenges we faced

Configuring the LLM to consistently produce high-quality recommendations was more complex than anticipated. It was a non-linear process, requiring significant trial and error to achieve meaningful improvements. Thankfully, we found great value in the abundant resources available to optimize LLM performance.

Accomplishments we’re proud of

We’re thrilled that our digital assistant not only handled all six hackathon prompts but also excelled with follow-up questions! Notable capabilities include:

Build a same-league team for VCT International, Challengers, or Game Changers
Build mixed-gender, cross-regional, and rising stars teams
Utilizing player statistical data to select a balanced team and provide evidence to the user of player performance
Identifying in-game leaders (IGLs) and player roles
Analyzing player agent pools and their strategic contributions
Assessing team strengths and weaknesses, while offering actionable improvement suggestions

Lessons learned

Through this project, we honed our skills in prompt engineering to refine LLM responses and implemented retrieval-augmented generation (RAG) to inject domain-specific knowledge. Additionally, we mastered deploying a robust, scalable infrastructure using CDK.

What’s next

We aim to enhance the assistant’s flexibility with more refined prompt engineering, expand its data sources for RAG, and explore integrating Agents to enable structured querying. Ultimately, we hope our digital assistant evolves into a powerful tool for managers and fans to better understand and engage with the VCT scene!