Turn messy raw data into balanced few-shot pairs

Transforms unstructured real-world operational logs or drafts into standardized input-output demonstration pairs. Use it to prepare reliable few-shot prompts.

You getScenario Summary, few-Shot Examples and edge Case Watchlist

Prompt

Prompt
Act as an AI evaluation specialist. Transform my raw operational data into three to five polished few-shot input-output examples for prompt design.

- Normalize tone, casing, and output fields across every pair.
- Select examples that cover distinct common scenarios, not duplicates.
- Strip sensitive internal identifiers while preserving realistic messy details.
- Ensure each output follows a consistent schema without missing keys.

Output format:
- Scenario Summary: one sentence per pair explaining the pattern it demonstrates.
- Few-Shot Examples: labeled clearly as Example 1, Example 2, etc., showing Input and Output blocks.
- Edge Case Watchlist: 2-3 inputs this set does not yet cover.

Here is my raw data:

Example output

Example for: raw support ticket notes about billing issues

Scenario Summary
  • Example 1 demonstrates an overcharge claim with an explicit invoice number.
  • Example 2 demonstrates an unclear refund request without transaction details.
  • Example 3 demonstrates a cancellation request caused by unexpected price tier changes.
Few-Shot Examples

Example 1
Input:
User says: "charged 49 bucks instead of 29 on inv-8812 help me out"

Output:
Category: Billing Error
Urgency: Medium
Extracted Invoice: INV-8812
Resolution Action: Verify tier pricing and issue partial refund.

Example 2
Input:
User says: "why is my card being billed again cancel this immediately"

Output:
Category: Cancellation Request
Urgency: High
Extracted Invoice: None provided
Resolution Action: Request account email and locate active subscription profile.

Example 3
Input:
User says: "noticed my plan changed to enterprise tier without me asking"

Output:
Category: Plan Dispute
Urgency: High
Extracted Invoice: None provided
Resolution Action: Check recent admin seat changes and audit plan upgrade logs.

Edge Case Watchlist
  • Multi-part requests where a user disputes a charge and requests account deletion simultaneously.
  • Foreign currency transactions with comma decimal separators.
Written by AI from this prompt. Your results will vary.

Paste your raw text, transcripts, or support tickets at the bottom of the prompt in ChatGPT, Claude, or Gemini. As a follow-up, ask the model to add one negative example demonstrating what the AI should refuse or flag.

Make it repeatable

Keep using this prompt without pasting it in every time.

Keep exploring

More in AI Tools & Prompt Engineering