Generate Dataset
The start_dataset_generation tool launches an asynchronous synthetic dataset generation pipeline on the Knowledgator platform.
Tool Overview
- Tool Name:
start_dataset_generation - Underlying Endpoint:
POST /datasets/generate/ - SDK Methods:
kg.datasets.generate(Python) /kg.datasets.generate(TypeScript)
Parameters
| Parameter | Type | Required | Description |
|---|---|---|---|
name | String | Yes | Name for the dataset (alphanumeric with hyphens/underscores). |
dataset_type | String | Yes | Service task type: ner, text-classification, classification, relation-extraction, or instruct-it. |
prompt | String | No | Task prompt describing data to generate (falls back to description/name). |
description | String | No | Optional description of the dataset purpose and prompt specifications. |
num_samples | Integer | No | Number of samples to generate (default: 100). |
domains | Array of Strings | No | Optional domain context tags (e.g. ["e-commerce", "finance"]). |
classes | Array of Strings | No | Target classification categories (for classification tasks). |
entity_types | Array of Strings | No | Entity tags to annotate (for NER tasks). |
json_template | String | No | Optional JSON schema template for structured generation. |
ner_output_format | String | No | NER annotation format (simple or tokenized, default: simple). |
Example Usage
Natural Language Prompt to Agent
"Generate 250 synthetic NER examples for e-commerce product reviews extracting 'product_name', 'rating', and 'defect_type'."
Tool Call Arguments
{
"name": "start_dataset_generation",
"arguments": {
"name": "ecommerce_reviews_ner",
"dataset_type": "ner",
"prompt": "Customer product reviews annotated with product names, sentiment ratings, and defects",
"num_samples": 250,
"domains": ["e-commerce", "retail"],
"entity_types": ["product_name", "rating", "defect_type"]
}
}
Tool Response
{
"pipeline_id": "e128a1c4-9844-4899-88c2-3e289bfca721",
"status": "pending",
"name": "ecommerce_reviews_ner",
"num_samples": 250,
"message": "Dataset generation pipeline started successfully."
}