QA/monitoring in recombinant vaccine manufacturing

On a16z[0] podcast on vaccines[1], an interesting tidbit came up: vaccine manufacturing in cell cultures is expected to hit a major challenge in terms of quality control due to unpredictable behaviour of cells.

The issue was mentioned only in passing during the podcast, specifically Rajeev Venkayya[2] said:

First of all, we need to see if these vaccines can be scaled up from a manufacturing standpoint very, very quickly. And not just be produced at scale but also produced with a reproducible high quality.
This is one of the biggest challenges that you run into when making biologics at large scale, particularly when we’re talking about hundreds of millions or even billions of doses.

I’m wondering how monitoring and quality assurance looks for cell cultures in an industrial setting. What are practical examples of what can go wrong with a cell culture that affect the final output (e.g. a vaccine) and how does one detect these issues? Is cytometry involved for "cell interrogation" or some more specialized methods are used?

I tried researching this question on my own, but quick googling didn’t yield much. I’m a software engineer with deep learning experience, I look at this issue from a computer vision point of view and wonder if anything my field could do to contribute to solving this issue.

[0] a VC firm with a science leaning
[1] which is very good! https://a16z.com/2020/08/14/vaccines-vaccinology-renaissance-covid-pandemic-beyond/
[2] board member of Coalition for Epidemic Preparedness Innovations