Training Example Feature

The Training Example feature allows users to improve the AI assistant by giving feedback on its responses. Admins can then review and curate the best examples, which are later used to guide the assistant’s behavior through the {{training_examples}} slot.

User Feedback Flow

Rate the Response

While chatting with the assistant, users can rate each answer:

• The response was helpful and accurate.
• The response was incorrect, confusing, or off-topic.

Add an Optional Comment

Users can provide a short explanation or additional context (optionalan). This helps reviewers understand why the feedback was given.

Better Comments = Better Learning

Thoughtful comments help reviewers determine how to adjust or improve the assistant’s answers.

---

Admin Training Workflow

All feedback is visible under the Training tab for reviewers.

Each row shows:

• The user question
• The AI answer
• The feedback or
• The comment
• Who submitted it and when

Admins can review and manage all submitted feedback in the Training tab. For each training example, the following actions are available:

• Edit: Improve or correct the assistant’s answer directly.
• Change Feedback: Update the user’s feedback (e.g., switch from to or vice versa) if it was incorrect or needs re-evaluation.
• Create Q&A: Promote a positively-rated example into a reusable Q&A pair, which is added to the assistant’s long-term knowledge base.
• Delete: Remove low-quality or irrelevant examples.

Feedback Editing

Q&A Creation

Q&A Conversion

When you click

to create a Q&A, it opens a prefilled form with the question and answer. You can modify them before saving. This new Q&A is stored in your data source and becomes part of the assistant’s core knowledge.

For more about how Q&A pairs work, see the Q&A Data Source

Keep it Clean

Changing feedback helps correct misclicks or adjust evaluations, ensuring only high-quality examples are used in training.

---

📚 Using Training Examples

The {{training_examples}} slot is used in prompt templates to provide the assistant with real-world examples of previous questions and answers. These examples help guide the assistant's behavior and improve its contextual understanding.

When this slot is used, the system automatically retrieves the 5 most relevant training examples based on the user's current question.

How Training Examples Are Added

• The example must then be manually confirmed by clicking the “Train” button in the admin interface.

Workspace Scope and Feedback Visibility

• Users with read & write access can view and manage all feedback submitted within the AI assistant.
• Users with read-only access can view feedback, but cannot edit or confirm it.
• If a user selects “Store privately” when creating the AI assistant, the entry is only visible to them.