valorant-scouting-report

• 

  Sample scouting report for FURIA

Inspiration

I am a big fan of Valorant as a Game, i play it a lot. I also watch VCT tournaments, i noticed that esports scouting is largely manual and inconsistent. I was inspired to explore how Agentic RAG could automate structured analysis the same way analysts think—by breaking scouting into strategies, tendencies, and compositions.

What I Learned

I learned how to design agent-based reasoning with RAG, manage real-world dependency conflicts, deploy on Hugging Face Spaces, and handle vector databases in ephemeral environments.

How I Built It

I built modular RAG agents using LangChain and ChromaDB, each responsible for a specific scouting dimension. A Gradio interface ties everything together and is deployed on Hugging Face Spaces via GitHub CI/CD.

Challenges

The biggest challenges were dependency conflicts, Python version mismatches, and handling non-persistent storage on Spaces. Solving these required rethinking versioning, startup logic, and vector indexing strategy.

What's next for valorant-scouting-report

With more time and support, I’m confident this system can deeply analyze and effectively break down opponent playstyles at a match and round level. Due to time constraints, I focused on building a solid Agentic RAG foundation rather than implementing full opponent-specific game-cracking logic.

Built With

• agents
• ai
• chroma
• crewai
• embedding
• langchain
• llm
• python
• rag