If you're fine training a model of your own, I would suggest try building a NER prediction model on your data. It would help to start with a certain type of data which follows similar pattern throughout. Once you're confident, you can extend it to many more patterns.
Or if you're lucky and find pre-trained NER models to be able to identify the event-date relationships already, you can just use them out-of-the-box.
Interestingly, I tried some QA models to simply ask the event & time of event given some context, and deepset/roberta-base-squad2 works pretty well for simple examples.
Context: I have a flight to catch at 6 AM tomorrow.
Question: What is the topic of event?
Answer: a flight to catch at 6 AM tomorrow
What is the date of event?
6 AM tomorrow
Context: The next set of features are scheduled to be discussed at meeting on 25th Dec 2022.
Question: What is the topic of event?
Answer: The next set of features
Question: What is the date of event?
Answer: 25th Dec 2022
Of course your data could be very different, but the pre-trained models should be a good starting point for ideas.