wow-alert

Vision-based World of Warcraft assistant — YOLO + OCR watches enemy cast bars and speaks the right callout, addon-free.

Python
YOLO
RapidOCR
PySide6
CUDA

Real-time cast-bar awareness for World of Warcraft, with no addons and no game hooks — pure computer vision. A PySide6 desktop app captures the WoW window, runs a YOLO detector to find enemy cast bars, OCRs the spell name, target and duration, matches the result against a curated per-dungeon spell database, and speaks the right callout (“KICK SPIRIT BOLT”) the moment a dangerous cast starts.

How it works

Detection: a custom-trained YOLO model finds cast bars in the captured frame in real time on a CUDA GPU.
Reading: RapidOCR extracts the spell name, target and cast duration from each detected bar.
Deciding: a rule engine matches spells against per-dungeon YAML databases (_global.yaml plus the active dungeon’s file, auto-selected by calibration), with severity levels and aliases for OCR fuzz.
Alerting: pre-rendered TTS phrases fire instantly — no synthesis latency mid-fight — while every detection streams into a live log pane.

Why it’s interesting

Game addons read game state through an API. This reads the screen, the way a human does, which makes it game-update-proof and genuinely general: the same pipeline (capture → detect → OCR → rules → alert) applies to any real-time UI you want to monitor.

More projects

argus LLM-as-optimizer for RAG pipelines — treats a retrieval cascade like a neural net and "trains" it with textual gradients until the loss plateaus.