Skip to content

Executors

Executors are the main interface for generating synthetic data with sintezi. They orchestrate formatting, API calls, parsing, and retry logic.

Available executors

StructuredAiCall

The primary executor for generating structured synthetic data from typed inputs to typed outputs.

from sintezi.ai.executor import StructuredAiCall, StructuredAiCallConfig, AiCallParameters
from sintezi.ai.formatter import auto_formatter_for_type
from sintezi.ai.parser import auto_parser_for_type

config = StructuredAiCallConfig(
    system_message="Your system prompt here",
    parameters=AiCallParameters(
        model="gpt-4o-mini",
        temperature=0.7,
    ),
)

ai_call = StructuredAiCall(
    ctx=ctx,
    config=config,
    formatter=auto_formatter_for_type(InputModel),
    parser=auto_parser_for_type(OutputModel),
    retry_policy=None,
)

result = await ai_call(input_data)

Key features:

  • Type-safe input and output with Pydantic models
  • Automatic format selection (JSON, XML, plain text)
  • Separate retry policies for network and validation errors
  • Support for loading configuration from files

Use cases:

  • Bulk synthetic data generation
  • Data augmentation (paraphrasing, translation, expansion)
  • Structured data transformation
  • Test data generation

Loading from files

Store prompts and parameters separately:

from pathlib import Path

ai_call = StructuredAiCall.from_files(
    ctx=ctx,
    path=Path("prompts/my_prompt"),
    formatter=auto_formatter_for_type(InputModel),
    parser=auto_parser_for_type(OutputModel),
)

This expects: - prompts/my_prompt.txt — system message - prompts/my_prompt.json — AiCallParameters as JSON

Future executors

More executors will be added in future versions:

  • Agent loop executor — for multi-step agentic workflows with tool use
  • Batch executor — for efficient bulk generation with OpenAI batch API
  • Streaming executor — for real-time synthetic data generation

Configuration

All executors share common configuration patterns:

AiCallParameters

from sintezi.ai.executor import AiCallParameters

params = AiCallParameters(
    model="gpt-4o-mini",
    temperature=0.7,
)

System messages

System messages define the LLM's behavior:

config = StructuredAiCallConfig(
    system_message="You are a data generator that creates realistic synthetic examples.",
    parameters=params,
)

Next steps