This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle
What I Built
League Coach is a stats-driven coach for League of Legends. You give it match stats (CS, deaths, vision, KDA, win/loss per game). It compares them to role benchmarks, finds your biggest gaps, and a local LLM turns them into a short coaching plan: your biggest problem, three drills for your next five games, and one thing you do well.
Who would pay: ranked grinders stuck in a rank who can't afford human coaching ($20-50/hour), and small coaches or Discord communities who want to triage students' stats before a session. Possible pricing: a free CLI, then a paid hosted version with Riot API auto-import and a weekly progress report.
Demo
Code
this is completely ai generated, I just wanted the sticker I am sorry :( it's supposed to be a league coach, at least I hope it works well
How I Built It
- Rules engine (pure Python, stdlib only): groups games by role and computes CS/min, vision/min, deaths in wins vs losses, and KDA. It ranks gaps against editable role benchmarks. The numbers come from code, not the model, so the LLM can't make up your stats.
-
Open-weight LLM via Ollama: the flagged gaps go to a local model (Llama 3.1 8B by default; Qwen 2.5 and others work via
--model), which writes the coaching under a strict "use only these numbers" system prompt. - Graceful fallback: if no model is running, you still get the rules-based findings.
Why Does Open Innovation Matter?
A coach is only useful if you run it after every session, so cost per run matters. With open weights it's free to run locally with no API key or rate limits. Your match history stays on your machine. And anyone can swap in a better model or edit the benchmarks for their own role, patch, or game, which a closed API wouldn't let me promise users.
Top comments (0)