Prompt2DAG: A Modular Prompting Approach for Democratizing Data Pipeline Generation

Building robust data pipelines often requires spe-cialized engineering skills, creating barriers for domain experts with limited coding expertise. We introduce Prompt2DAG, a modular prompting methodology that transforms natural language descriptions into executable Apache Airflow workflows by decomposing generation into three sequential stages: structured analysis, configuration generation, and code implementation. This approach aligns with established software engineering principles of separation of concerns and progressive refinement. Our evalu-ation across five different LLMs demonstrates that Prompt2DAG significantly outperforms conventional end-to-end generation, im-proving code quality (+78.4 %) and structural integrity (+43.2 %) of generated pipelines. Using a data enrichment case study, we show how this approach enables the development of high-quality workflows through natural language, effectively democratizing data pipeline development.