What are few-shot examples?
Few-shot examples are sample scenario outputs that are included in the generation prompt to guide the AI’s format, style, and quality. They demonstrate what good output looks like so the LLM produces consistent, well-structured scenarios.How they work
Each few-shot example contains the same three fields as a generated scenario:| Field | Purpose |
|---|---|
| User Perspective | Example of a well-written caller description |
| Expected Outcome | Example of step-by-step agent behavior |
| Intent | Example of a clear intent statement |
OUTPUT FORMAT EXAMPLES. The AI uses these as templates for formatting and structure.
Global vs. ring-specific examples
Few-shot examples can be scoped two ways:| Scope | When used |
|---|---|
Global (ring_id = null) | Used for all rings, as fallback |
Ring-specific (ring_id = N) | Used only when generating for that specific ring |
Agent type filtering
Examples are also filtered by agent type:| Agent type | Which examples are used |
|---|---|
| Inbound | Examples marked inbound or both |
| Outbound | Examples marked outbound or both |
Managing few-shot examples
Few-shot examples can be managed through the SuperBryn UI. Each example has:- The three text fields (user_perspective, expected_outcome, user_intent)
- A display order controlling priority
- An agent type filter (inbound, outbound, or both)
- An optional ring ID for ring-specific scoping
Best practices
- Provide at least 1-2 global examples to ensure every ring has guidance
- Add ring-specific examples for rings where output quality needs to be different (e.g., Ring 2 policy scenarios look very different from Ring 5 conversational scenarios)
- Keep examples representative of your domain — use real scenarios that passed quality review
- Update examples as your understanding of good output evolves