LLM-Driven Open Banking and SAP Integration Framework: A Scalable Cloud Architecture using Databricks AI, Gradient Boosting, and Automated Software Testing

Abstract

In the rapidly evolving financial ecosystem, open banking initiatives demand agile, intelligent and secure architectures that can integrate legacy enterprise systems with modern AI-driven services. This paper proposes a scalable cloud framework that leverages large language models (LLMs) for understanding and orchestrating open banking APIs, integrates with the enterprise resource planning backbone of SAP S/4HANA (and related modules), utilises the Databricks Lakehouse platform for unified data and AI workloads, and employs gradient boosting machine (GBM) models for structured-data predictive tasks. In addition, the framework embeds automated software testing pipelines to ensure reliability, compliance and continuous delivery. The architecture supports real-time or near-real-time scenarios such as account linking, consent management, payment initiation, fraud monitoring and regulatory reporting. It allows an LLM interface to parse natural-language requests (e.g., from fintechs or corporate clients) into open banking transactions and maps them into SAP-centric business processes. The Databricks Lakehouse abstracts the data ingestion, transformation, feature engineering and model serving layers; the GBM component handles high-volume structured-data tasks such as credit risk scoring and anomaly detection; and the automated testing ensures the integrity of API integrations, model updates and end-to-end workflows. We describe the conceptual architecture, component interactions, design criteria (governance, latency, scalability, security, explainability), and implementation methodology. The proposed framework addresses both business agility and operational resilience, offering advantages in responsiveness, reuse, and governance, while discussing limitations around complexity, model drift and regulatory risk.