Building KRXY — Part 6: Two Raspberry Pis, One Network — What Actually Gets Faster?Part 5 ended with three hosted Qwen2.5 deployments, a model library, and evidence that choosing a model was about more than its parameter count. But the network itself still had only one physical Rasp6h ago · 10 min readkrxy.hashnode.dev
Building KRXY — Part 5: Measuring Answers Before Choosing a ModelPart 4 ended with a gateway that could send a request to the exact hosted model a node reported, refuse an incompatible route, and avoid turning a repeated request into two jobs. The next question was1d ago · 10 min readkrxy.hashnode.dev
Building KRXY — Part 4: When a Model Name Stops Being EnoughPart 3 ended with a Raspberry Pi that could identify itself, report its health, and make a signed claim about what it could do. The next question was whether that claim should decide where a job went.3d ago · 7 min readkrxy.hashnode.dev
Building KRXY — Part 3: Giving a Raspberry Pi an Identity in a Decentralised AI NetworkPart 2 ended with a machine that could answer a question. The Hailo runtime was on the Pi, bound to loopback, and the control plane on the Mac could hand it a job. That was a working accelerator. It w4d ago · 11 min readkrxy.hashnode.dev
Building KRXY — Part 2: Building the First Raspberry Pi AI NodePart 1 was the plan. The diagrams were coherent, and the hardware list was specific enough to order from. None of that answered the first practical question: could a Raspberry Pi 5 and an AI accelerat6d ago · 15 min readkrxy.hashnode.dev
