Jetson Orin Nano powers local voice assistant

Good idea in spirit, but only if you treat it as a hobbyist edge-AI build, not a polished Alexa replacement. The hardware is capable enough for small local models, but the budget and software friction will decide whether this feels like a clever gift or a weekend project that keeps growing.

• –NVIDIA now positions the Orin Nano line for edge generative AI, with the Super dev kit pushing 8GB LPDDR5 hardware to 67 TOPS and up to 25W/MAXN modes, so the compute story is real.
• –8GB RAM is still tight for anything beyond small quantized models; 3B-4B class models are the sensible ceiling if you want responsiveness and room for speech stack overhead.
• –`llama.cpp` on Jetson is viable, but community issues show the usual Jetson pain points: CUDA/build quirks, GPU-driver detection problems, and model-specific regressions that make setup less plug-and-play than x86.
• –For a gift, the higher-probability win is a complete, pre-assembled voice appliance with a known-good mic array, speaker, NVMe storage, and a tested speech pipeline rather than a bare board plus open-ended model experimentation.
• –If the recipient already runs a home server, the cleaner architecture may be a local voice front-end on the Jetson and the heavier orchestration on the server over LAN.