Current methods for large scale tandem mass spectrometry analysis are brittle. These methods are also unsuited for general or untargeted applications. Unlike image or text data humans cannot easily use their general knowledge or intuition to label mass spectra, which makes accumulating large scale datasets even harder. I aim to use self supervised learning techniques (view based like MoCo, SimCLR, RoPAWS, mixup etc.) to train a encoder capable of generating general and useful embeddings for MS2 spectra acquired in glycomics experiments. Once I am able to verify the quality of these embedding and the encoder my aim is to finetune/train a layer on top of frozen encoder weights. This finetuning layer will be designed for de novo annotation of glycan spectra with proposed glycan structures.