Category: Research & Experiments
-
This is a note on a design decision behind IRIDIUM ONE rather than a finished result, why the hardware baseline uses AMD’s ROCm stack instead of the far more common Nvidia/CUDA path most local-AI projects default to. The Conventional Default Most local-AI tooling in the wild, Ollama’s own documentation included, is written with an Nvidia…
Search
Latest Posts
- Research Note: Betting On ROCm Instead of CUDA for On-Device Inference
- Grounding Local Models Without Retraining: The RAG Pipeline Behind IRIDIUM ONE
- How CODR Classifies a Coding Task Before Routing It Anywhere
- Inside IRIDIUM CODR: How a Multi-Provider Coding Dispatcher Actually Routes Work
- Inside IRIDIUM ONE: Architecture of a Privacy-First On-Prem AI Appliance