Low-code data engineering environments increasingly support rapid data ingestion, transformation, and integration, yet schema matching remains a persistent challenge when heterogeneous datasets use inconsistent naming conventions, limited metadata, and weak documentation. This study proposes a lightweight explainable framework for schema matching and data fusion in low-code data engineering pipelines. The framework integrates sentence embedding similarity, fuzzy string matching, keyword-based heuristics, and data-type compatibility validation to identify candidate correspondences between independently structured tabular schemas without requiring labeled training data, predefined ontologies, or extensive manual mapping. The approach is evaluated using publicly available Airbnb Listings and Reviews datasets, where the Listings file contains property-level attributes and the Reviews file contains review-event attributes, creating a practical case for schema alignment across related but differently structured data sources. In addition to hybrid similarity scoring, the framework includes an optional large language model based justification layer that generates natural-language explanations for predicted column matches, improving interpretability for analysts, data engineers, and low-code users. The experimental results show that a moderate similarity threshold of 0.65 achieved a precision of 0.50, recall of 0.33, and F1-score of 0.40, while a stricter threshold of 0.80 achieved a precision of 1.00, recall of 0.25, and F1-score of 0.40. These findings demonstrate the trade-off between broader mapping coverage and high-confidence matching, while confirming the value of explainable scoring for low-supervision schema alignment. The proposed framework contributes a domain-agnostic and low-footprint method for preparing heterogeneous data sources for downstream data fusion, ETL automation, and analytical workflows in low-code data engineering settings.