Inspiration

Built for VCT coaches and IGLs who need instant opponent analysis. Tired of manual VOD review and static scouting reports, I wanted an AI system that delivers actionable tactical intel in seconds - strategies, player tendencies, comps, and counter-plans.

What it does

Valorant ScoutMaster analyzes any VCT team (NRG, LOUD, Sentinels) and delivers instant scouting reports + tactical counter-plans. Enter a team name → get strategies, player tendencies, compositions, and exploitable vulnerabilities with executable IGL calls.

How we built it

Dual AI Pipeline: RAG Scout Engine - 3x ChromaDBs (all-MiniLM-L6-v2) extract team strategies, player habits, comps from curated datasets TacticalPlanner Crew - 3 Gemini 3 Flash agents (Studying/Planning/CounterPlanner) generate deep analysis Gradio UI - Clean web interface with real-time progress bars HF Spaces - Production deployment with secrets management

Challenges we ran into

GRID API dead-ends: GraphQL limits, missing round events → built custom semantic datasets CrewAI Knowledge bugs: OpenAI embedding forced despite Gemini LLMs → stripped Knowledge, used explicit inputs ChromaDB corruption: Rust panic in Docker → rm -rf vector_db* + fresh rebuilds 120+ dependency hell: HF Spaces timeouts → 18-line minimal requirements.txt

Accomplishments that we're proud of

Zero-config deployment - HF Spaces live in 2 minutes Production RAG - 3 specialized vector DBs > framework defaults Agent creativity - Temperature progression (0.8→1.2) unlocks pro-level tactical thinking

What we learned

Custom RAG > framework Knowledge (CrewAI's is broken) crew.kickoff(inputs={}) > context inheritance magic Version-less requirements.txt = HF Spaces bulletproof Gemini 3 Flash crushes esports reasoning (1M context) Progress bars = professional UX polish

What's next for Valorant-Tactical-AI

Live VLR.gg match scraping Real-time map-specific counter-calls Voice IGL commands ("NRG Bind default") VCT playoff predictor Team-vs-team matchup simulator Mobile PWA for LAN scouting

Built With

  • agentic-ai
  • agentic-rag
  • ai
  • chromadb
  • crewai
  • gemini
  • gradio
  • langchain
  • python
  • rag
  • sdk
Share this project:

Updates