In this study, we propose a system that uses a depth camera and skeletal tracking to detect poor posture during desk work and alerts users. This approach eliminates challenges associated with sensor attachment and installation, typically faced by conventional systems that rely on multiple sensors, by assessing various postural issues using a single sensor. The system focuses on the alignment of core body parts such as the head and torso and evaluates posture based on misalignment features. An Intel RealSense depth camera and NuiTrack were used for skeletal tracking, whereas an LSTM-based posture classification model was implemented to detect common poor postures, such as slouching, sacral sitting, and scoliosis, from the tracking data. As the camera’s field of view is limited to the upper body, the system leverages the time series capability of the LSTM model to effectively detect poor posture even without lower-body data. Initial experiments showed that slouching and sacral sitting could be classified with reasonable accuracy. However, the detection of scoliosis remains challenging. Future work will focus on improving the model with more diverse data and developing a real-time warning system that visually displays a user’s core body misalignment.