Prompt’s Evolution for Language Model-Driven Data Generation

In the era of data, data generation continues to grow day by day, posing new challenges for processing systems due to the dynamism of its volume, variety, and velocity. To process data in real time, stream processing systems (SPS) serve as the keystone of real-time data analytics systems. However, SPS must operate under strict quality of service (QoS) constraints, which require dynamic adaptation of their internal logic to sustain performance. In this work, we address the problem of automatic text data generation using large language models (LLMs) by employing an evolutionary approach to guide prompt learning. The process enables us to automatically discover prompts by applying a black-box approach to generate synthetic data from a reference dataset. This approach aims to enrich training datasets and enhance the generalization capabilities of AI-driven adaptive SPS models.