A maker known as CreativelyBankrupt has built Sparky, a fully offline suitcase robot powered by an Nvidia Jetson Orin NS Super 16GB. The robot runs a large language model locally and uses over 30 sensors to interact with its environment.
Sparky runs entirely on the Jetson module
Sparky runs Gemma 4 E4B at Q4_K_M via llama.cpp with q8_0 KV cache and flash attention. It supports a 12K token context and achieves roughly 200 ms time to first token, with an inference speed of about 14-15 tokens per second. For speech, it uses SenseVoiceSmall for speech-to-text and Piper for text-to-speech, with mouth animation updating at 43 Hz.

The project demonstrates what is possible with current edge AI hardware. By running everything locally on a Jetson module, Sparky avoids cloud dependency and keeps all processing on the device. The combination of a capable LLM, multiple sensors, and synchronized facial animation makes for a responsive and expressive robot.
CreativelyBankrupt shared details of the build with Tom's Hardware. Pricing and availability for Sparky have not been announced, and it remains a custom project rather than a commercial product.



Discussion
0 comments
Log in to join the thread with a thoughtful take, question, or correction.