This paper presents a novel framework for estimating upper limb joint angles by synchronizing surface electromyography (sEMG) signals with kinematic data acquired from a motion capture system. A custom Python-based program was developed to coordinate data collection, ensuring that recordings from both the sEMG device and motion capture system commenced and terminated simultaneously. Long Short-Term Memory (LSTM) networks were employed as the predictive model for joint angle estimation. sEMG signals were recorded from three muscles, including the triceps brachii, biceps brachii, and deltoideus medius, using the Shimmer3 EMG Unit. This device features two sEMG channels and streams data wirelessly to a computer via Bluetooth. Joint kinematic data were concurrently captured using the Vicon motion capture system, which employs high-speed cameras to provide ground truth data for training the LSTM model. Participants performed a set of upper limb movements, including movements at the elbow (flexion and extension), shoulder (abduction, adduction, flexion, and extension), along with free motions. To enhance signal quality, the EMG data were filtered using band-pass and notch filters, and time-domain features such as variance and root mean square (RMS) were computed for input into the model. The proposed LSTM model demonstrated adequate performance, achieving a Mean Absolute Error (MAE) of approximately \(8.2\% \, \pm \,3\%\) of the total range of motion for each joint. This method establishes a potential framework for recognizing motor intent, offering strong potential for integration into assistive technologies such as robotic exoskeletons.