NUC AI Model Strategy (Sprint 4)

Objective

Stress test the Intel NUC 13 Pro (Orchestrator Node - i5) with a comprehensive suite of "Builder" and "Designer" models to evaluate routing capability, headless stability, and CPU-optimized performance (Alderlake) before offloading creative tasks to the Mac Mini M4 Pro.

Model Stack (Ollama)

1. The Builders (Coding & Technical)

• deepseek-r1:32b: Heavyweight reasoning and logic (19GB).
• deepseek-r1:7b: Primary fast reasoning.
• qwen2.5-coder:7b: Specialized code generation and refactoring.
• llama3.1:8b: Stable general-purpose baseline for technical documentation.
• codellama:7b: Legacy coding support and specialized Python/C++ tasks.

2. The Designers (Analysis & Rendering Ops)

• llava:7b: Multi-modal vision support for analyzing system screenshots and UI layouts.
• medusa:latest: High-throughput inference for rapid-fire prototyping.

Performance Metrics (TBD)

• Inference Speed: Tokens per second on CPU (optimized via Alderlake/OpenVINO backend).
• Memory Pressure: Impact on 64GB system RAM during multi-model orchestration.
• Routing: OpenClaw capability to switch between specific models for "fit for purpose" task execution.

Deployment Status

• [x] Initial Stack Pull initiated (2026-03-02).
• [ ] Thermal and stability audit.
• [ ] Integration with topics/ documentation workflow.