Skip to main content

Generate Dataset Plan

The generate_dataset_plan tool generates a structured dataset creation plan (taxonomy, sample schema, and distribution) from a task description prompt.


Tool Overview​

  • Tool Name: generate_dataset_plan
  • Underlying Endpoint: POST /datasets/generate/plan/
  • SDK Methods: kg.datasets.generate_plan (Python) / kg.datasets.generatePlan (TypeScript)
Generation on the Platform vs. Managed Agent Remediation
  • Use generate_dataset_plan to synthesize brand new datasets from natural language prompts via the platform.
  • To audit, rebalance, and repair an existing dataset on disk, use Knowledgator's on-platform Managed Agent. See Dataset Remediation & Skills.

Parameters​

ParameterTypeRequiredDescription
promptStringYesDetailed prompt describing the goal, target domain, and schema for the dataset.

Example Usage​

Natural Language Prompt to Agent​

"Generate a dataset plan for training an NER model to extract financial terms, revenues, and dates from corporate 10-K filings."

Tool Call Arguments​

{
"name": "generate_dataset_plan",
"arguments": {
"prompt": "NER dataset for financial 10-K filings extracting revenues, operating expenses, and dates"
}
}

Tool Response​

{
"status": "success",
"plan": {
"domain": "Finance / Regulatory Filings",
"suggested_types": ["revenue", "operating_expense", "filing_date"],
"distribution": {
"revenue": 0.45,
"operating_expense": 0.35,
"filing_date": 0.20
},
"guidelines": "Ensure entities capture surrounding currency symbols and full calendar dates."
}
}