Why I Took Data Science Courses as an Infrastructure Engineer
My MSCS at CU Boulder included a graduate certificate in Data Science. Probability theory, statistical inference, regression and classification, data mining methods and pipeline. On paper, that has nothing to do with my actual job.
In practice, it shows up more than you’d think.
SLO math
The first time I tried to set an SLO for platform availability, I realized I didn’t actually understand what “99.9% available” meant in statistical terms. How many data points do you need before that number is meaningful? What’s the confidence interval? If you measured 99.95% over 7 days, is that statistically different from 99.9%?
These aren’t academic questions. They determine whether you alert your on-call engineer at 3 AM or not. They determine whether you tell leadership the platform is meeting its targets or missing them. Get the math wrong and you’re either waking people up for nothing or sleeping through real incidents.
The probability and inference courses gave me the vocabulary and the math to think about reliability metrics properly. When I set an error budget now, I can explain why the window is 30 days and not 7, and why the burn rate threshold is where it is.
Logs
Figuring out why deploy times spiked last Tuesday? That’s data mining. The capacity planning spreadsheet I keep for our clusters is basically regression. Even incident categorization is classification if you squint at it.
I was doing all of these things before the coursework. But I was doing them by eyeballing spreadsheets and going with my gut. The data mining courses gave me actual tools, and more importantly, they helped me recognize when my gut was wrong.
The graduate project
One of my graduate projects involved building a classification pipeline for detecting PII in text data. That specific use case isn’t relevant to my day job. But the process, acquiring data, cleaning it, selecting features, training a model, evaluating results, iterating, is the same process you use for any data-driven decision.
The data mining project course forced me to take a problem end-to-end. Not just “here’s a clean dataset, run a model.” Real data. Messy data. Data that required judgment calls about what to keep and what to throw away. That’s closer to what production observability data looks like than any textbook example.
Am I a Data Scientist?
No. I’m an infrastructure engineer who can read a paper and not be lost, and who knows enough statistics to tell when a metric is real and when it’s noise. I don’t need to build neural networks. I just need to set alert thresholds that actually mean something.
If your MSCS program offers data science electives, they’re worth considering even if you’re on the infrastructure track. The overlap isn’t obvious until you’re knee-deep in SLO dashboards wondering if your numbers actually mean anything.