Value Driven Data Science: Boost your impact. Earn what you’re worth. Rewrite your career algorithm.

[00:00:00] Dr Genevieve Hayes: Hello, and welcome to your value boost from Value Driven Data Science, the podcast that helps data scientists transform their technical expertise into tangible business value, career autonomy, and financial reward. I'm Dr. Genevieve Hayes, and I'm here again with Andrei Oprisan. Head of engineering at agent.
[00:00:21] ai to turbocharge your data science career in less time than it takes to run a simple query. In today's episode, we're going to explore how data scientists can leverage the current AI boom to accelerate their career progression. Welcome back, Andre.
[00:00:40] Andrei Oprisan: Thank you. Great to be here.
[00:00:41] Dr Genevieve Hayes: So as I mentioned at the start of our previous episode together, we are at the dawn of an AI revolution with unprecedented opportunities for data scientists.
[00:00:51] Now, through your current role at Agent. ai, and prior roles at AI centric companies, such as OneScreen. ai, you've clearly managed to capitalize on this AI boom, and are actively continuing to do so, and have managed to build a very impressive career for yourself, partly as a result. Now, the Internet's full of career tips, but they're usually very generic advice from career coaches who've never worked in the data science or technology space, and their advice usually doesn't take into account the specific context of the AI landscape.
[00:01:35] What's one specific strategy that data scientists can use right now to leverage the AI boom for faster career progression?
[00:01:44] Andrei Oprisan: I would say first building some expertise and prompt engineering and AI model evaluation. I think that's a foundation on top of that. I think it's developing some systematic approaches for comparing different models outputs on domain specific tasks and then creating something maybe like a reliable evaluation framework.
[00:02:05] For example, you could create an eval set. Or tasks in a field and developing some quantitative or qualitative metrics to assess how different models perform compared to traditional approaches and that can really position you as someone who can actually properly integrate AI tools into existing workflows while having that element of scientific rigor.
[00:02:28] , it's leveraging the existing trends around prompt engineering around the different models that are coming up every week, every month. Every quarter and figuring out, how we are going to showcase when to maybe use 1 versus another with the scientific approach with again, I would start as simple as.
[00:02:47] An eval from the kind of work that you're doing in your current role or organization, or thinking about adjacent organizations and adjacent kind of strategies to then create some examples of when and when you wouldn't. Use certain models because of, some numbers where you can show in an email that, this model does really well in this kind of let's say, classification in this specific domain versus. One that doesn't . I think from there, you can iterate and do some even more interesting work very repeatedly and looking at some adjacent domains and apply the same sort of technical solutioning to other domains.
[00:03:26] Dr Genevieve Hayes: I read an article recently that was written shortly after the launch of the DeepSeek LLM. And there was a group of researchers at a university that were evaluating the model. And they had a series of prompts that could be used to find out, can this model be used to produce offensive or dangerous information?
[00:03:49] And they had something like 50 prompts and they randomly chose 10 of them and ran it against that. Is that the same sort of thing that you're proposing, but obviously specific to the person's organization?
[00:04:03] Andrei Oprisan: That's exactly it. So I think starting as simple as again this prompt engineering and writing out a few of those prompts and be able to get some kind of repeatable answer, whether it's a score, whether it's, selecting from a set of options, just anything that you can then repeat and measure in a Quantitative way
[00:04:24] and like, we can say, okay, it is this category, we're getting with these, let's say 50 prompts we're consistently getting, 10 percent of the answers are incorrect, but 90 percent where we're getting this kind of consistent answer and an answer that can actually be useful.
[00:04:40] And then looking at different kinds of models and and then figuring out, how do they form? But also, how might you improve that? And apply some level of scientific method thinking around, ultimately, what can you change to improve? Essentially, what are still these for most folks, black boxes these LLMs that, And go something outcome, something else, and maybe demystifying what that looks like in terms of consistency at the very least in terms of accuracy over time.
[00:05:12] And then, it could even take on more advanced topics. Like. How can you improve those results once you have a baseline starting point, you can say, okay, sure. Now, here's how I improved, or here's how maybe the prompts were. Incorrect or, they behave differently given a different LLM or, maybe you push different boundaries around context window size on the Google models are not the best.
[00:05:38] But they're the best at dealing with large data sets. there's a trade off at a certain point in terms of speed and accuracy and cost.
[00:05:47] And so then introducing some of these different dimensions, or maybe only looking at those in terms of, you know, yes, if this LLM takes 10 seconds to get me a 98 percent accurate answer, but this other one takes half a second to give me a 95 percent accurate answer, which one would you choose and a business context essentially the faster one that is a little bit cheaper.
[00:06:11] Might actually be the right answer. So there's different kinds of trade offs, I think, given different kinds of context. And I think exploring what that might look like would be a really good way to kind of apply some of those technical skills and looking at some of those other dimensions, around things like pricing and runtime execution time.
[00:06:31] Dr Genevieve Hayes: And I can guarantee if you take a strategy like this, you will become the AI expert in your office, and you will be invited to every single AI centric meeting the senior management have forevermore because I did something similar to this it was before LLMs. It was with those cloud cognitive service type APIs.
[00:06:50] And anytime one of those came up, I was the person people thought of. I got invited to the meeting. So, this is really good career advice.
[00:06:59] Andrei Oprisan: And really, it starts, I think, growth especially think about how do you grow your career as a technical person? Obviously, part of it is being in the right room at the right time to be able to ask the right kinds of questions to be able to present a technical perspective. And again, I think by pushing on some of these boundaries you get exposed to even bigger.
[00:07:20] Opportunities and bigger challenges that do need technical solutions that do need someone with a technical mind to say, You know what? Maybe that doesn't make sense. Or maybe there is a way to leverage a I, for this problem, but not maybe in the way that you're thinking, and I think being able to at least present that perspective is incredibly valuable.
[00:07:39] Dr Genevieve Hayes: And regardless of which industry you're working in, the secret to success is you've got to get in the room where it happens, as the Hamilton song says, and this sounds like a really good strategy for getting there with regard to LLMs.
[00:07:53] That's a wrap for today's Value Boost, but if you want more insights from Andre, you're in luck.
[00:08:00] We've got a longer episode with Andre where we discuss how data scientists can grow into business leadership roles by exploring Andre's own career evolution from technology specialist to seasoned technology leader. And it's packed with no nonsense advice for turning your data skills into serious clout, cash and career freedom.
[00:08:23] You can find it now, wherever you found this episode, or at your favorite podcast platform. Thanks for joining me again, Andre.
[00:08:31] Andrei Oprisan: for having me. This is great.
[00:08:33] Dr Genevieve Hayes: And for those in the audience, thanks for listening. I'm Dr. Genevieve Hayes, and this has been Value Driven Data Science.