Active · May 2026
Fully Offline Edge AI Assistant
A complete speech-to-speech AI stack — recognition, reasoning over a local knowledge base, and synthesis — running on embedded hardware with zero network connection.
Ask it a question out loud, and it answers — with no internet, no cloud, and no data leaving the box.
This project packages an entire conversational AI pipeline onto a single embedded edge device: speech-to-text, an LLM reasoning layer with retrieval over a local knowledge base (including a full offline encyclopedia), and text-to-speech, all running locally.
Why it matters
Disconnected environments are the norm, not the exception, in the places we care about: field sites, secure facilities, vehicles, and eventually spacecraft — where round-trip light lag makes cloud AI physically impossible. An assistant that works with the network cable cut is a fundamentally different tool from one that merely tolerates a slow connection.
Status
Operational internally and under active development — current work focuses on latency, knowledge-base tooling, and agent capabilities on-device.