Is it possible to create a named entity recognition system without using POS tagging in the corpus?
Yes, both of them are different tasks. POS tagging helps NER systems but it is not necessary. You can get features (say BERT/ELMo embedding) for each word in the sentence and train a CRF NER model. This looks like simple example https://www.pragnakalp.com/bert-named-entity-recognition-ner-tutorial-demo/