CursorAI/Forum/Post/24/02/My Best Practices for MDC rules and troubleshooting

• https://academy.langchain.com/courses/take/intro-to-langsmith/lessons/60631062-lesson-2-annotation-queues

• LangSmith Annotation Queues overview

  • can have feedback applied to them
  • can be individually added to a dataset
    • #Q when should we encourage SMEs to add to a distinct LangSmith/Dataset vs when to apply a #Rubric ?
      • The annotation queue has a "D" hotkey for adding to the default dataset, so it might be simplest if we were going to only use datasets for the "golden" dataset - the "reference" of a good response.

• Creating LangSmith Annotation Queues

  • Selecting Default LangSmith/Dataset

    • in the example, they select RAG Application Golden Dataset

  • Creating LangSmith/Annotation/Rubric

    • we can leave high level instructions, e.g. "Please only score fields you are comfortable with"
    • full list of feedback tags from our LangSmith/Workspace
      • See also
        • Set up feedback criteria | 🦜️🛠️ LangSmith
          • https://smith.langchain.com/settings/workspaces/feedbacks -> link to edit feedback for workspace
        • Feedback data format | 🦜️🛠️ LangSmith
          • this is the data format for each "feedback item"
            • continuous
            • categorical
              • could be yes/no (binary)
    • For each field we can define special instructions specifically in the context of this annotation #Cool
  • Reservation Locks
    • ensures that multiple people don't provide feedback for the same run at the same time

• Going through the Annotation Queue