Indoor localization technology that leverages Wi-Fi has become the preferred choice for researchers, primarily due to its cost-effectiveness and ease of deployment. However, the presence of indoor obstacles introduces a certain level of volatility in the RSSI (Received Signal Strength Indicator). To mitigate this, the implementation of deep learning techniques to process the collected RSSI signals is practical. This paper presents a multi-task deep learning framework that is adaptable to various buildings and floors. The framework incorporates a joint classification method for buildings and floors, along with regression analysis based on longitude and latitude coordinates. Shared encoders are employed to decrease the number of parameters and reduce computational costs. Experiments were conducted using the UJIIndoorLoc dataset. When balanced weights were applied to both classification and regression tasks, the classification accuracy reached 99.72%, and the best mean two-dimensional location coordinate error was 8.81 m.