Hey — I'm Matthew, a mechanic from Saskatoon. This isn't a company or a product. It's a side project I started because I wanted a real Jarvis-style companion: something I can just talk to, that can see what I see and speak back naturally. Her name is Winter.

**The architecture**

My gaming PC (Ryzen 7 7800X3D, 64GB RAM) is the local "suit computer" — it runs the whole loop. The pipeline is streaming speech-to-text → AI brain → streaming text-to-speech, fully pipelined and interruptible. I can cut her off mid-sentence and she stops and listens — measured stop time around 0.3 seconds on real speech. Warm response latency (end of my sentence to first audio back) averages about 1.1 seconds. If she mishears me, she asks me to repeat instead of going silent.

Honest bit: the brain is currently a cloud AI (Anthropic), not local inference. Running the brain on my own PC is the dream, not the reality yet. Everything else — voice loop, cameras, glasses bridge — runs on my own hardware ove