Business decision making at Kitabisa.com depends on information held in the corporate data warehouse. Operational teams still struggle to retrieve that data themselves because they lack the technical background, leaving daily requests queued with the data team and allowing business terms to be read differently across divisions. This research designs, implements, and evaluates a conversational artificial-intelligence agent built on Retrieval-Augmented Generation. Regression testing closed at 39 cases, all passed. Across 24 conversational scenarios, 20 analytical questions were answered and 4 risky requests refused. All eleven SQL security rules held, and masking withheld identifying columns. A web evaluation of ten scenarios recorded a mean end-to-end latency of 17,323 ms across a range of 5,063 to 28,564 ms, with every scenario reaching its expected terminal state without manual SQL correction. A seven-item usability assessment with two operational-team respondents scored 4.43 of 5.00, or 88.57 percent of the maximum, indicating that the prototype is fit for first-level analytical exploration by the operational team.