In software engineering, the testing phase is critical to ensuring application quality, reliability, and adherence to defined requirements. Effective testing requires realistic, consistent data that reflects real-world conditions and exists in sufficient volumes. This makes the data seeding process indispensable. However, traditional approaches to data creation—manual or automated—face challenges such as ensuring data diversity and compliance with business rules, often resulting in inadequate test coverage with consequences in undetected bugs, reduced reliability, higher costs and compromised software quality. Thus, the goal of this paper is to present a novel generative AI-based architecture for advancing data seeding practices in software testing. The proposed architecture integrates user interaction, data embedding, and retrieval-augmented generation (RAG) to form a seamless pipeline for generating contextually relevant and realistic data in real-time. For the purposes of validating the architecture, a Proof of Concept (POC) was implemented focused on generating fake data from Brazilian customers to assess the quality of the data generated and the effectiveness of the implemented solution, in order to evaluate the potential of the architecture for the development of broader applications and in different contexts.