Unified AI compute router — LLM inference, image, video, TTS/STT, OCR — across 6 GPU backends with auto-failover.