Our senior design project with r4 Technologies focused on making business data easier to access for non-technical users. At r4, valuable enterprise data already existed, but answering business questions often required experience with internal analytics tools. This created a barrier for users who needed insights quickly but could not directly interact with the data. To address this, we designed a natural language interface that allows users to ask business questions in plain English and receive answers generated from approved enterprise data. Our system used Vertex AI Gemini models in a two-pass pipeline. In the first pass, the model interpreted the user’s question by identifying intent, extracting key information such as filters and metrics, and creating a structured query plan. In the second pass, the model generated SQL from that plan rather than directly from raw language. This approach improved reliability, control, and debuggability compared to one-step generation. Because the project involved enterprise data access, safety and trust were major priorities. We added guardrails so that only approved schema elements could be used, only read-only SQL queries were allowed, and all generated queries were validated before execution. The final result was a system that reduced the technical burden of querying business data while maintaining accuracy, structure, and security.