Ask HN: How do I reliably eval my AI models

I created an AI agent and have been considering using Braintrust or Arize - but where can I source experts who actually can make goldsets for me too? Is it worth sourcing them?