Using machine learning in geophysics is often considered as a fast approach to process or interpret seismic data, but the challenge is to get enough data to train the machine learning core. This framework uses a combination of real noise data and synthetic reflection seismograms generated from e.g. real source signal or band-limited pulses for training the machine learning core. The trained core can be stored on (IoT) devices which can be used in the field to preprocess the data, for e.g. QC, before sending it to the office for further processing. This will decrease the turn-around time and will help geophysicists to decide whether the data is useful for further processing or needs to be re-collected. I will explain the framework, discuss the results, and show how the framework improves the seismic data quality. The framework can deconvolve the seismic data to zero-phase band-limited pulses with simultaneous noise reduction.