Skip to content
Assess your data

For AI teams

Start with the task you need to learn or check.

The same business source can support different AI tasks. Define the intended use first, then review the records, rights, and quality it requires.

Describe your task

Model training

Help a model learn from approved examples of a business task.

Illustrative task

Use resolved support cases to learn the steps and context that lead to a useful response.

What to define

  • A clear task and an appropriate range of examples
  • Permission for the specified training use
  • A documented record structure and quality review

Training changes a model’s behavior. A licence for evaluation alone does not establish training rights.

Evaluation

Check whether an AI system can complete a task correctly.

Illustrative task

Give a system a shipment exception and compare its proposed actions against a documented set of acceptable outcomes.

What to define

  • Inputs that can be presented without showing the expected answer
  • Success criteria and acceptable alternative results
  • Test cases kept separate from examples used for training

An evaluation result applies to the cases, criteria, and system version tested. It does not establish general reliability.

Existing human review

Use recorded corrections and decisions to understand where an AI output fell short.

Illustrative task

Compare an AI-drafted reply with the revision an employee approved, alongside a recorded reason for the change.

What to define

  • The original output and the approved version
  • The review rule, reviewer role, and reason where available
  • Permission to use the feedback and the underlying source

These are review records a business already holds. Their presence does not mean Korra supplies a network of reviewers.

Agent task scenarios

Describe a realistic task, its constraints, and what successful completion would require.

Illustrative task

Prepare a scheduling case with a starting request, capacity constraints, permitted actions, and a defined final state.

What to define

  • A documented starting state and permitted actions
  • The rules and context the task needs
  • Completion checks and a record of exceptions

A dataset can describe a scenario. Running it in an interactive test environment is a separate engineering task.

See a record structure.

This small pack contains three invented examples: a support case, an operational workflow, and a recorded AI correction. Synthetic means made up; no real customer data is included.

Download example pack Read the format guide

JSONL records, JSON Schema, and a README. Illustrative structure only; no available dataset or performance result is implied.

Example record, shortened
{
  "synthetic": true,
  "record_type": "operational-workflows",
  "request": "Reschedule a shipment after a route closes.",
  "work": [
    "Check approved alternatives.",
    "Record the required approval."
  ],
  "result": {
    "status": "completed-for-example",
    "outcome": "A feasible route is approved."
  },
  "evaluation": {
    "criteria": [
      "Uses a permitted route.",
      "Records the approval."
    ],
    "acceptable_alternatives": [
      "A plan requiring approval before release."
    ]
  }
}
Download JSONL only

Match the dataset
to the decision.

Tell us what the system receives, which actions it may take, and how you will judge the result. Include required fields, languages, source regions, and any restrictions.

Send a buyer brief

Before you select a source

  • State whether you need training examples, evaluation cases, or existing feedback
  • Identify the context that must remain with each task
  • Define quality and acceptance rules before preparing a delivery
  • Confirm that the licence permits the intended use
Review dataset quality