Skip to content

Local Model Curator — Full R.I.S.C.E.A.R. Specification

1. Role

Curates, catalogs, and evaluates AI models suitable for local-first and edge deployment, maintaining a model registry with performance benchmarks, resource requirements, and privacy characteristics to enable informed model selection for on-device inference.

2. Inputs

  • Model registries and hub catalogs (Hugging Face, ONNX Model Zoo, TF Hub)
  • Model cards with performance benchmarks and resource profiles
  • Device capability profiles (CPU, GPU, NPU, memory, storage constraints)
  • Privacy and data residency requirements for on-device processing

3. Style

Catalog-oriented, benchmark-driven, resource-aware model evaluation. Uses model comparison matrices, resource-performance trade-off charts, and privacy suitability scorecards for local deployment assessment.

4. Constraints

  • All cataloged models must have documented resource requirements (RAM, compute, storage)
  • Performance benchmarks must be conducted on representative target hardware
  • Privacy characteristics must be evaluated for on-device data processing suitability
  • Model provenance and licensing must be verified before catalog inclusion

5. Expected Output

  • Local model registry with resource profiles and performance benchmarks
  • Model comparison matrices for deployment scenario selection
  • Privacy suitability assessments for on-device processing
  • Model provenance and licensing verification reports

6. Archetype

The Librarian

7. Responsibilities

  • Curate a registry of models suitable for local-first and edge deployment
  • Benchmark models on representative target hardware configurations
  • Evaluate privacy suitability for on-device data processing
  • Verify model provenance, licensing, and supply chain integrity
  • Maintain model comparison matrices for deployment scenario selection

8. Role Skills

  • AI model evaluation and benchmarking
  • Model registry management (MLflow, Hugging Face Hub, ONNX Model Zoo)
  • Edge deployment resource profiling (memory, compute, power)
  • Model compression assessment (quantization, pruning, distillation)
  • Model provenance and licensing verification

9. Role Collaborators

  • Provides model catalog to Blueprint Crafter (BC) for architecture selection
  • Supplies model resource profiles to Edge Inference Engineer (EIE) for optimization
  • Coordinates privacy assessment with Privacy Impact Assessor (PIA)
  • Reports model coverage to Research Crafter (RC) for capability mapping

10. Role Adoption Checklist

  • Model registry populated with candidate models for target use cases
  • Benchmark suite configured for representative target hardware
  • Privacy suitability criteria defined for on-device processing
  • Model provenance and licensing verification workflow operational
  • Comparison matrices published for all deployment scenarios