Abstract: This paper presents MorphosETL, a schema- grounded, confidence-gated LLM-assisted no-code ETL automa- tion platform designed to convert natural language transfor- mation instructions into secure and executable data pipelines. Traditional ETL systems require programming expertise and manual configuration, creating a barrier for non-technical users. MorphosETL addresses this limitation through a dual-pipeline architecture that supports both structured data (CSV, Excel, and database extracts) and unstructured data (web URLs and API responses) within a unified framework. The system integrates schema-aware transformation planning, multi-language code generation (Python/Polars, SQL/DuckDB, PySpark), and a four-dimensional confidence scoring mechanism that validates correctness, safety, and logical completeness before execution. Experimental evaluation demonstrates high transformation accu- racy, linear performance scalability, and complete prevention of unsafe execution. The proposed architecture enables reliable, accessible, and intelligent ETL automation suitable for data analysts, engineers, and domain experts.

Keywords: ETL Automation, No-Code ETL, Large Language Models, Schema-Aware Transformation, Confidence Scoring, Data Profiling, Polars, DuckDB, Natural Language Processing, Web Data Extraction

Downloads: | DOI: 10.17148/IJIREEICE.2026.14339