How we tune local models for enrichment and connect them as custom provider inference through Orchestration Lab.