← All projects
AIGames

LeagueCoach

A real-time coach that talks you through a League of Legends match.

LeagueCoach

The problem

Decisions in League happen in seconds, and coaching usually arrives afterward in a VOD review. By the time you know you should have backed off, the fight is over.

The approach

LeagueCoach polls the Live Client Data API every few seconds for the full game state, captures the minimap and template-matches every champion icon to find who is where, and tracks how long each enemy has been unseen. That state is built into a structured prompt, and the coach fires on events (first blood, an objective taken, your death, enemies missing too long) and on a timer. Advice is spoken aloud.

AI architecture

Claude receives a structured game-state block: phase, your status, teammates, enemies with missing-time warnings, minimap positions mapped to named regions, recent events, and estimated objective spawn timers. The system prompt encodes objective priority, wave management, and phase-specific priorities, and asks for short, actionable advice. Text-to-speech delivers it.

Highlights

  • ▹Fuses two data sources: Riot's Live Client Data API and OpenCV minimap detection
  • ▹Missing-enemy tracking from last-seen timestamps, with gank-path warnings
  • ▹Event-driven triggers: first blood, objectives, deaths, enemies MIA
  • ▹Objective spawn timers estimated from the kill history
  • ▹Spoken delivery so the player never looks away from the game