Ryan Connelly

I'm developing a fully local, modular AI framework that empowers users with persistent memory agents, optimized for consumer-grade hardware. My focus areas include: Local-first AI: Stage 1: Deploying models like Mistral Nemo Instruct 2407 in ONNX format, optimized for AMD Ryzen AI NPUs. (80% complete) Stage 2: Deploying models from the Mamba 2 family in ONNX format, optimized for AMD Ryzen AI NPUs. (65% complete) Multi-agent orchestration: Crafting collaborative agent systems with shared, persistent memory and recursive reflection loops. Ethical AI: Ensuring user autonomy and privacy without reliance on cloud services. Open-source contributions: Sharing tools and models to foster community-driven innovation. I'm passionate about democratizing AI access, and building systems which respect user agency and provide localized functionality equal to much larger models/systems.