Large Language Models (LLMs) are shaping new directions in automated pipeline creation, where workflows can be described and built directly from natural language prompts. Pipelines are essential for tasks such as data processing, machine learning model training, and cloud-based workflows. Existing approaches often depend on rigid templates, which restrict adaptability, increase manual workload, and limit scalability when applied to diverse or dynamic tasks. Without adaptive configuration, such pipelines remain inefficient for real-world conditions. This work introduces a prompt-driven pipeline synthesis framework with context-aware auto-configuration. The method interprets prompts to design pipelines that automatically select modules, adjust execution parameters, and structure components according to the task context. Reinforcement learning (RL) is applied to refine decision-making, while graph-based optimization enhances interconnections between pipeline stages. This hybrid approach supports automation, maintains adaptability, and scales effectively across different workloads. The framework is evaluated on case studies involving machine learning workflows, data preparation, and deployment tasks. Key metrics include task accuracy, adaptability score, runtime efficiency, configuration overhead, and intervention rate. Results demonstrate that the proposed method achieves higher accuracy, lower reconfiguration costs, and better adaptability compared to template-based approaches. The findings confirm that context-aware auto-configuration guided by LLMs can create intelligent pipelines that combine automation, performance, and scalability. This study contributes a practical foundation for adaptive pipeline synthesis in modern computing environments.