Continuous mode delivery of medical oxygen from oxygen concentrators and oxygen cylinders leads to wastage of precious medical oxygen during exhalation and rest phases of the respiratory cycle. Pulse mode oxygen delivery can address the stated problem, however, it is required to determine the number of oxygen release pulses and the exact instant of inhalation or exhalation. Herein we report the design and development of an intelligent pulsed mode respiratory device- “RESPIPulse,” which is capable of delivering oxygen bolus by automatically sensing the inhalation and exhalation instances from body mount surface electromyography (sEMG) electrodes without manual intervention or settings. The device comprises a set of miniature single-channel sEMG electrodes, an embedded machine-learning algorithm, a normally open solenoid valve, an airflow sensor, and necessary driving electronics. The solenoid valve opens or closes depending on the muscular inhalation or exhalation effort determined from the sEMG signals, thus preventing the wastage of respiratory oxygen. The sEMG signals are subjected to envelop extraction followed by feature extraction. Performances of k-nearest neighbor (kNN), support vector regression (SVR), and random forests (RF) regressors are initially tested in Python IDE to identify the best learning algorithm that is deployed in the microcontroller for determination of the instances of inhalation and exhalation. Trials are conducted on 20 healthy subjects and 10 dyspnea-affected patients. Based on the computed performance measures and evaluation time, the kNN algorithm estimates the respiratory instances more accurately than the other two algorithms. A significant amount of oxygen savings, ranging between 35.48–82.35%, is obtained using the RESPIPulse device which is much higher than the pulse mode delivery devices employing manual settings exhibiting maximum conservation of 48.2%.