I worked on the problem of resume parsing for a year and a half. Pretrained language models helped us generate tons of Named Entity Recognition models that do a pretty decent job finding entities! This worked out pretty great for cities, names, businesses, universities. The real hard part is piping all the NERs to groupers that yield accurate groupings. Sometimes your NER misses a school but has a degree, related project descriptions and a graduation date. You have to re-run the NER on that area and make a best guess or give up. Lots of interesting applied NLP engineering problems there! I gave up though. If someone has more patience, I think this approach can really improve things.