Hugging Face's blog published a build write-up by Rick Grandy, an NVIDIA contributor, describing MonsterHouse, an interactive Halloween installation that turns a garage door into a cranky, pirate-themed character. Trick-or-treaters speak to the house, which replies in the voice of a grumpy old ship built from the timbers of a sunken vessel. Its wooden mouth is animated in Unreal Engine 5 with virtual fire behind the door. The house brushes off rude visitors, but a polite "please" gets through immediately and "thank you" earns a warm goodbye. The project runs on a single workstation with two NVIDIA RTX PRO Blackwell GPUs, according to the author. The system chains several NVIDIA open models. Nemotron Speech handles listening and turns speech into text. Nemotron 3.5 Bolt E4B, a small open language model with a LoRA adapter for personality, writes each reply. Magpie TTS Zeroshot speaks the reply in a voice cloned from a short reference sample, and Audio2Face-3D converts the audio into mouth movement. Nemotron 3 Nano Omni describes costumes, while YOLO and ByteTrack count and track visitors. The author says NVIDIA's Nemotron Voice Agent supplied the conversation foundation, which left the physical character rig, built in Autodesk Maya, as the most labour-intensive part. The author chose a self-contained, off-grid setup because of memory and compute limits, and says the character keeps a memory of visitors who have already won him over. For Australian businesses and professionals, this changes little directly. MonsterHouse is a personal project, not a product or service. It is still a practical reference for anyone prototyping a local voice agent with open-weight models, and the hardware cost is far above consumer levels. Any public deployment that records the voices or images of children or visitors would need to be assessed against the Privacy Act 1988 and the Australian Privacy Principles, even if the processing stays on site.