LLM-Driven Decision Automation: Integrating SQL and Java APIs through Multi-Step Reasoning Pipelines

The rapid evolution of Large Language Models (LLMs) has enabled new paradigms in intelligent decision automation by transforming how systems interpret, reason over, and act upon both structured and unstructured data. By integrating LLM chains with structured data systems such as SQL databases and enterprise-grade Java APIs, organizations can automate complex decision-making workflows that traditionally required human intervention, domain expertise, and manual rule configuration. These integrations allow LLMs to not only generate context-aware queries but also execute business logic through API orchestration, effectively bridging the gap between natural language understanding and operational systems. This paper proposes a unified architecture that combines prompt chaining, retrieval-augmented generation (RAG), and programmatic API orchestration to enable scalable and reliable decision automation across heterogeneous enterprise environments. By leveraging iterative reasoning, contextual grounding, and feedback-driven refinement, the approach enhances accuracy, reduces hallucination, and improves explainability in automated decisions. The study synthesizes recent advancements in text-to-SQL systems, LLM orchestration frameworks, and API-driven enterprise systems, demonstrating how multi-step reasoning pipelines can be operationalized in real-world environments to support use cases such as dynamic reporting, intelligent workflow execution, anomaly detection, and real-time decision support while maintaining robustness, scalability, and enterprise compliance requirements.