People & AI - Bridging AI and Pathology with Julianna Ianni

In this episode, Julianna Ianni from Proscia discusses the transformative power of AI in medical imaging and pathology with host Karthik Ramakrishnan. They explore the importance of diverse data for model development, the global applicability of AI models, and the exciting future of AI in accelerating diagnosis and improving accuracy. Julianna emphasizes the need to understand business aspects and improve communication skills for researchers. They also discuss Proscia's innovations in digitizing pathology slides, automated quality control, and balancing innovation with practicality. The conversation highlights regulatory challenges, the impact on healthcare accessibility, and advice for women in the AI industry.
June 7, 2024
5 min read

Listen to the full episode now!





Karthik Ramakrishnan [00:00:07]

Welcome to the People and AI podcast, where we explore the transformative power of AI across industry. I'm Karthik Ramakrishnan, and today we have a special guest who is at the forefront of AI and pathology. Julianna Ianni, vice president of AI research and development at Proscia, is here to share her insights on real world use cases of AI in pathology, best practices for developing models, and her inspiring career journey. Julianna, let's start with your background. Can you share with us the backstory of how you decided to pursue a career in AI, particularly in the field of pathology? What drew you to the field and keeps you excited about its future?

Julianna Ianni [00:00:46]

Yeah, absolutely. I think I've always had an interest in sort of health technology and biotechnology. In high school, I spent a lot of meal times, like, reading news articles about the latest of these technologies. So I knew I liked technology, and I knew I wanted to help people. AI and pathology came to me, I think, a little later. I did an internship in biomedical informatics during my undergrad at Vanderbilt University, and I think what I took away from that is just, wow, this medical data is powerful. But I also was thinking, everyone's looking at other medical data, imaging has so much more of it, and I just. I knew I needed to learn about imaging next.

So I pursued my PhD in biomedical engineering doing MRI research. And during my PhD, I was developing algorithms solving optimization problems for image reconstruction and for tailoring RF pulses for image acquisition. So, in layman's terms, just getting the best data for an image and putting together the best image once you have that data. I wasn't really doing machine learning per se until the latter half of my PhD, but it came really easily at that point, because I had this informatics background in machine learning, I think is just a special case of solving an optimization problem. And I was already familiar with that. I think, though, the deep learning part, while it was during my PhD, that I was kind of seeing that take off, and the MRI research community was a little slow on the uptake, or maybe more likely, I was kind of impatient. But I knew. I knew deep learning and medical imaging was really about to take off, and I felt like I needed to be in on the ground floor.

And luckily, that's when I found the folks at Proscia, who introduced me to the field of pathology, which was, I think, even further behind MRI on the uptake, because it was coming from a place where most pathology slides, these slides of literally your tissue biopsies, they weren't digitized. And Prosha had this vision of bringing AI to pathology to help patients, and I was sold on that challenge and the promise of doing that. So, yeah, that's how I came to join Proscia and build out the AI team here.

Karthik Ramakrishnan [00:03:27]

So AI, and I wouldn't say, but broadly in healthcare and medicine, is making some big waves. I would love for you to kind of illustrate some of what you think is one of the coolest real world applications of AI and pathology. And maybe we can start with the breakthrough that just happened, or the new, latest breakthrough that happened with deepMinds, alpha fold models in drug discovery. It would be great if you could break down how that actually happens. What was the breakthrough? And what did DeepMind do with alphafold to make that come to life?

Julianna Ianni [00:04:09]

Yeah. Yeah. So alphafold, I think, continues to amaze all of us. But the latest happening is that they released this new model, and it not only predicts the sort of protein structure, which is what the original alphafold was able to do, but now they're able to predict how these biological molecules will interact. So predicting how DNA, rna, and how small molecules can interact with each other, these things all used to be incredibly hard problems and still are hard problems to solve. Taking at least weeks, if not months, solve this sort of thing. And for some things, you might never really solve it. So, just amazing that they could do this.

It's like having instructions to put together a really, really hard, really, really expensive Lego kit before you buy it. Except these Lego constructions are drugs that can help patients. I think this sort of capability is going to have an enormous impact on drug discovery. We're already seeing the impacts of other models accelerating that process. Aside from drug discovery. Even though pharmaceutical organizations contract research organizations, they're already developing and using AI in drug development as well. I think, like nearly every other industry, they're benefiting from AI, and AI is sort of poised to really impact this one and bring treatments to patients faster at basically every step of the process. I think we're not far from seeing the first drugs that owe their existence to AI sort of go through approval and make their way to patients.

The impact of that is going to be just tremendous.

Karthik Ramakrishnan [00:06:17]

Right. But maybe we would bring it back to Proscia if you could comment on some of the most interesting projects you and your team are working on. What keeps you excited?

Julianna Ianni [00:06:26]

One of the cool things that we're working on right now is an automated quality control application for some context. Pathology is important in both research and diagnosis of diseases like cancer. And the labs conduct research in drug discovery and development can play a big role in shaping future breakthrough therapies. But they're working with thousands or hundreds of thousands of tissue samples in the form of these digitized images. Those images, though, can have a lot of quality issues, just like the ones that you might take on your phone. They might have out of focus regions, or in particular, when we're dealing with tissue, that tissue might be folded over on itself. So there's all sorts of things like that that can happen. So it's not unheard of for scientists to spend weeks analyzing images.

I'm sure something like this has happened to you before, and then later find out that the data has some major quality issues. So our automated quality control application uses AI to find these issues. As soon as those images are scanned, rather than either missing them entirely or having someone have to manually scan through all of those images, it sort of can take a long time for someone to do that. It's sort of like looking for one street on a really big map because these images are so large. So we automate that process.

Karthik Ramakrishnan [00:08:02]

And what's the impact of these advancements that you're bringing at Proscia in patient care and research in the field?

Julianna Ianni [00:09:10]

Yeah, I think maybe it sort of makes sense to start with a little bit of context there, because pathology has. Has sort of traditionally relied on glass microscope slides, and you obviously can't apply AI on these slides until they've been digitized. So in addition to building AI, Proscia offers a platform called concentric that diagnostic laboratories and life sciences organizations use for all aspects of their digital pathology operations. And when I look at the impact of what we're doing at Proscia, there are a few examples to highlight. First, our core platform helps to drive confidence and efficiency gains simply by just enabling the pathologist to work with the images in the first place. And then that can result in faster turnaround times for patients, for example, meaning they can possibly start treatment sooner. And that's really critical because there's an intensifying shortage of pathologists. There's fewer and fewer pathologists every year, and cancer rates continue to rise globally.

So that's an area where not only our platform, but also AI applications can make an impact. And by helping these organizations to go digital to digitize their pathology slides, we're also enabling them to create a whole bunch of data that can be used for AI development. And our platform also offers developer tools for organizations to do this. I also talked about our automated quality control module. It's specifically for life sciences R&D teams right now, and enables them to QC their images up to six times faster than manual review alone and at high power in a time trial. In turn, they can start studies faster and work with higher quality data, which means that research and whatever they're working on, whether it's developing a drug or something entirely different, all of that is higher quality because the data going into it is higher quality. I think more generally, we like to say at Proscia that AI can solve 10,000 problems in pathology. So I could talk about that for quite a while.

Karthik Ramakrishnan [00:10:36]

No, that's fair. Maybe we'll get back to at least a couple of those where it could be applied. The thing about accessibility of these systems and the speed at which you can do these, obviously in highly populated countries, I think this could be meaningful because there's a distinct deficiency in the number of medical professionals that are available on a per capita basis. And I think that's if you can bring down the cost of medical care, I think that's another real business benefit that comes out, but also at the bottom line for an individual. Right. Like from there, what they have to pay out of pocket, I think those are just kind of daily impacts that we can see, not just from a, the fact that we can detect diseases or faster, but it's also the cheaper aspect of it that I think is pretty critical. So when you're thinking about these newer way you're advancing, how do you balance the innovation that you want to do versus the practicality of productizing the model? Right. The model in and of itself doesn't really do much right.

You have to put the product around it. So how do you guys think about how your end customer consumes that model? So are you building custom models? Is it a standard model that people can use out of your developer tools? Walk us through that process. How would one use Proscia?

Julianna Ianni [00:12:04]

Yeah, so we definitely spend a lot of our time thinking about how somebody is going to use the model that we're developing. That is a whole lot of the work, maybe more of the work, than developing the model in the first place. In terms of balancing innovation and practicality, I personally don't feel a tug between the two. I know that a lot of people do. And sort of in the big picture, the world needs some people focused on the innovation alone. But I've personally never really been interested in innovation for innovation's sake.

Karthik Ramakrishnan [00:12:43]

There you go.

Julianna Ianni [00:12:44]

I want to help people today. I want to help people yesterday. And I think that my colleagues at Proscia share that passion. And to me, that means building AI that matters. It means building AI that's going to make it directly into the hands of people who need to use it to help patients. And I'm saying that from a position of sort of privilege, as someone who works in digital pathology today, that I think these problems require innovation. They just require it. We get to innovate with immediate purpose.

And does that mean we never spend time cleaning data? No. Does that mean we never pull a model off the shelf? No. Like, absolutely no. We pull whatever we can pull. That, I think, is what puts models in a place where they can help patients. And there are just so many pieces that require innovation and creativity to do it, both in terms of building the models themselves, but also in terms of building all the infrastructure and the integrations to deliver them.

Karthik Ramakrishnan [00:13:52]

Yeah, absolutely. I think that's it. I think OpenAI really opened our eyes to that, because we've had models in AI for a decade or more. I mean, longer than a decade, but even the transformers have existed for a long time, that they really showed how it can be made consumable at scale by the layperson. I think that's really what the productization of that simplicity of that productization. Right. And the delivery, I think that had a bigger impact than the model itself for all those years. Right? Yeah, absolutely.

And I see what you mean. And so you talked about the models that you pull off the shelf or, you know, or are you building custom algorithms? What types of models and algorithms are using? Obviously working with vision based models from the image processing that you're talking about. But what's your favorite model? Even if I can go there.

Julianna Ianni [00:14:53]

What's my favourite model? Do I have a favourite model? I don't think I have a favourite model. What will get the job done? And there's a lot of different models we explore. We work with vision models. We work with language models. It's a little bit. It's a little bit of everything. Sometimes we're building our own, sometimes we're not.

I think, yeah, it's a lot of variety. I think we're also exploring ways to accelerate development with foundation models. I think there's a lot of exciting things happening there, and it's super cool to be in a place in AI, like at this moment in time where all this, all this technology is exploding. I feel like we're really lucky to witness that. But I couldn't tell you what's my favourite.

Karthik Ramakrishnan [00:15:43]

But the reason I understand I was speaking to someone in the banking industry and data scientists, they work with structured data. So you're working in the unstructured world. They work in the structured world, tabular world. And he's like, my favourite is the XGBoost. I don't need a deep neural network or generative AI model because XGBoost just does the job and that's my go to every single time. And, yeah, they'll enhance it, of course, and put a bunch of different things in. That's why I just had that conversation last night with someone. So I was just curious, how would you say this whole generative AI, sort of the transformers and the gender of AI breakthrough, if you will, if you want to call it that. How's that impacted your work and how you're able to do more now?

Julianna Ianni [00:16:29]

Yeah, I think a lot, but I think the most is yet to come, I guess, in terms of generative AI, I think we're all just starting to figure out the ways that we benefit from it. I think personally, I benefit a lot from ChatGPT. That helps. That helps me with my everyday to like quite a bit. But we're also, I think, you know, going to see other, other benefits from some of these generative AI models.

Karthik Ramakrishnan [00:17:07]

Yeah, I mean, from a productization or a product development standpoint for you guys. Yeah, I'm sure the way that you can work with multimodal aspects of data in an unstructured context, I think must have been a trigger. I don't know if it was really. I mean, I guess my question really is, was there an inflection point or was it business as usual? You were already using them because you were aware of them.

Julianna Ianni [00:17:34]

I would say that there was probably an inflection point, at least for me. I don't know if everyone else at Proscia would agree, but it seems like to me, everyone has had a personal inflection point where it's like we were all, you know, we all looked at, we looked at Bert differently, right? Like, we knew, we knew Bert's limitations. And those of us who especially who are not, like Proscia, has historically been a vision company, right. In terms of our AI models. So those of us who were like, kind of asleep on the, on the language bit, uh, at some point looked around and said, hey, whoa, this is different.

And for me, that was ChatGPT coming up.

Karthik Ramakrishnan [00:18:25]


Julianna Ianni [00:18:25]

I don't know what it was for everyone else.

Karthik Ramakrishnan [00:18:28]

Yeah, fair enough. Yeah. I mean, oh, yeah. Let's talk about Bert another day. Okay. So, you're using models. You're applying this in the medical sphere, specifically pathology. What are some of the best practices that you've come across into developing and deploying AI models in a field as critical as pathology?

Julianna Ianni [00:18:52]

Yeah, absolutely. I think depending on the application, pathology, a lot of the rules here are similar to what they are in every other industry. And in terms of best practices, what you need to do. And so for us, one of the things that's really important is having good data to start with. So one of, you know, one of the things that we need to pay attention to is first having good quality data, like having the right labels, whether those are slide level labels, image level labels, or whether those are like, annotations has drawn, for example. You just have to start with clean data, and that makes 90% of the difference, I think.

Karthik Ramakrishnan [00:19:44]


Julianna Ianni [00:19:46]

And then beyond that, really understanding the domain and that ground truth. For example, we were developing this melanoma application previously that when we really got, when we really got into it, we started to understand, okay, our ground truth isn't really like truth. There's no objective truth. And I think this is true in more domains than people realize. Melanoma, for example, you may not know, pathologists disagree on a diagnosis a good bit out of the time, and it's not the most severe cases of melanoma that there's a disagreement on some of those earlier stage tumors. You could ask an opinion from two different pathologists and you'll get two different answers. And so we found in that we had to be very careful about what we considered to be our truth and what we were telling our models.

Karthik Ramakrishnan [00:20:45]

Yeah, we can take this a whole other level of what's objective through this truth, even objective. But in this case, I can see that can be quite difficult. Right. Like how one physician would treat or consider a certain type of outcome or the diagnostic of a certain, I guess, a picture or an image of a cancer versus another, they can approach it very differently. And they could both be right. And I think the treatment plasma would create could be different, but they could all. Both lead to the same outcome. So how do you pick? Is it a coin toss or is it variability? You could say, well, in the spectrum of things, here's the five options you could try. Right. That have worked.

Julianna Ianni [00:21:29]

Yeah. Well, I think there's many ways to approach that. The way that we chose to approach that, and we've published some research on this at the time, was to basically say, okay, this is not a binary problem. This is a score, essentially.

Karthik Ramakrishnan [00:21:52]


Julianna Ianni [00:21:53]

And so we got the opinions of several different pathologists for each of the cases that we were trading on. We asked. That's what we asked of our model, was to give us a score. How many of these pathologists would say, this is melanoma versus benign, but many, many ways that you could approach that problem.

Karthik Ramakrishnan [00:22:17]

Good. And then going back to data, this is. I mean, no data, no AI, and having good data, there's no substitute having good data. I mean, that's it. That's the starting point. And I think you're reiterating that, of course.

Now switching gears a little bit, obviously, healthcare of highly regulated industry and space, how do you think of the regulatory challenges or the regulatory environment under which you operate and how do you navigate them?

Julianna Ianni [00:22:49]

Yeah, yeah. So for. I think for Proscia and for pathology in general, there's sort of several different environments that we operate in in terms of regulations. So I think first is the clinical environment, which is probably the one that you think of the most, and that's in the US, regulated by the FDA. As far as medical devices go with pathology, I feel like the FDA is still learning and we're still learning with them. This stuff is still new, especially in pathology. Unlike radiology, the way that FDA has approached pathology image analysis, in the way that regulators globally have approached this, as it's something, it's a new category. So you can't go just get a 510 clearance based on an AI application.

In radiology, pathology is a whole new ballgame. And that's because the pathology images have not been digitized for decades. So the digitization in pathology is still relatively new on the scene. And that means that these AI applications are relatively new on the scenes. It's a little bit behind radiology in that regard, but catching up to a large extent. But that just is, it means it's a whole learning process. So in addition to going through the sort of standard challenges in medicine and AI, we have an added layer. But then I think on the other side of things, even outside what's in the past been considered a regulated medical device.

AI systems targeted at patients, whoever is developing them, are now subject to an increasing amount of regulation. So that's a new challenge that the industry is dealing with right now. I think the last category of regulation that sometimes is less discussed. Not disgusting, just to be clear on that.

Karthik Ramakrishnan [00:25:06]

I heard you.

Julianna Ianni [00:25:09]

But we talked about the drug development side of things. That's actually a regulated business as well, in terms of the FDA regulates that through what's called GxP requirements, and that can have indirect implications for both the functionality and development of the types of AI solutions that we build and that others build in pathology.

Karthik Ramakrishnan [00:25:38]

Yeah. And feel free to. Not to answer this question, but, you know, do you think, how do you feel about the regulations? Are they moving in lockstep? Are they moving in the right direction? And if you had to give the regulators any feedback, what would you tell them right now?

Julianna Ianni [00:25:57]

Oh, that's a scary question.

Karthik Ramakrishnan [00:26:01]

I can give you another question, but I would love an answer.

Julianna Ianni [00:26:04]

I will give an attempt at an answer to that. So I already told you, Karthik, that I'm personally, I'm impatient.

Karthik Ramakrishnan [00:26:12]


Julianna Ianni [00:26:13]

So that's. I think that's my answer is there's no way, there's no way that any regulatory body could move fast enough to make me happy on this stuff because I want to get this stuff into the hands of patients, into the hands of people helping patients. I think that's probably the best answer I could give you there.

Karthik Ramakrishnan [00:26:33]

So you just want them to move faster. And that's good, because I think the ambiguity that can, in and of itself, forget the regulatory policy in and of itself, it's just the sheer act of being slow can also be stifling to innovation, too, because you live in sort of flux.

Julianna Ianni [00:26:53]

Yeah. It's almost an impossible a job. You want to get these innovations out and get them helping people, but you also want to make sure they're really helping people. And that's incredibly tough when you have the technology just changing under your feet.

Karthik Ramakrishnan [00:27:13]

Yeah, good. And one of the things also I think about a lot in healthcare is the diversity of our population has to be reflected in the diversity of the data that you have. Right. Because racial backgrounds, genders, the way diseases express themselves, can vary quite a bit by race as well. So how do you think about that as you're developing these solutions? And how do you think about the diversity of the data that you need?

Julianna Ianni [00:27:47]

Yeah, that's something that I think about a lot, and that's been, it's been studied in medicine and even in digital pathology in particular. I think it's doctor Oakton Rader is somebody whose publications I recommend looking up on that if you're interested in that topic, in medical imaging, some really cool stuff that they've done. But just illustrating this problem of like, if you develop with the wrong data, if your data is not diverse enough, you really see some negative impacts. And so that's something that we take a lot of care to do, is develop with as much diverse data as we can possibly get our hands on. But it's certainly a challenge for everybody.

Karthik Ramakrishnan [00:28:41]

Right. And do you think, and maybe just an extension of that same question? I'm just wondering if you developed a model based on data in the US, and let's say you're a client of yours in let's say, India. Right. Or Korea for that matter. Right. Would they be able to leverage the same model or would you need to retrain it with the specificity of that population spectrum?

Julianna Ianni [00:29:05]

I don't think I can answer that. Just like, in a broad sense, very much depends on the type of data and the nuances of the data or the images, in this case, that you're running on. And also very much depends on how the model is trained and what it's doing. Sometimes those models can work in that situation. Sometimes they've been, sometimes they've been explicitly and very aggressively trained to work on a wide variety of data and the task is suited for it. Sometimes not. Sometimes you may need to retrain that model on the specific data, or at least fine tune that model. Data from a specific site.

Karthik Ramakrishnan [00:29:54]

Fair. Yeah, totally acceptable. It depends answer, right, because you need to first test and does it work? And if it doesn't work, why isn't it working? You had to find a solution for that. No, I get it, but it's, you know, I wish there was a general answer, but that never is. Right.

Julianna Ianni [00:30:14]

But I wish there was a general answer for that too, because that. Then we've all solved the generalization problem and everyone can go home.

Karthik Ramakrishnan [00:30:22]

Yeah, what the hell? Generative AI needs to step up its game, make it more generalized.

Okay, so rapid fire, a few questions. So obviously, the AI industry is evolving quite quickly. I mean, the kind of shifting under our feet every week, I feel like just in terms of all the things that are happening. So what are the things that most excite you about the future of AI and maybe the future of AI and pathology specifically, even.

Julianna Ianni [00:30:49]

Yeah, AI is moving fast in, I feel like every industry these days, it's so impossible to keep up. I'm most excited about the medical applications, obviously, things like adding new treatments to patients faster. In pathology specifically. I can give you another example of work that we've done at Proscia. For example, we published research on a system that can categorize skin cancer cases. I also talked a little bit about our research on melanoma and predicting to what degree pathologists will agree on that diagnosis. So those are just a couple examples of ultimately how AI might be used in pathology. So things like accelerating diagnosis, improving diagnostic accuracy, identifying patients who are probably going to be good candidates for a specific drug or therapy, those are all possibilities that are really exciting to me, and some of them are already being realized.

Karthik Ramakrishnan [00:31:54]

Very cool. And that's, I think, key. We love to stay in touch as you see most advances and bring you back to talk about them. But at the same time, what are you concerned about in the way the industry is moving?

Julianna Ianni [00:32:09]

Yeah, I think one thing that I'm concerned about is organizations having a sort of a false start to the generative AI revolution. In this particular moment, I think we're seeing a lot of excitement among organizations now, particularly around generative AI like we discussed. And I think those that were sort of snoozing on AI's potential have seen now what's possible, and they kind of want in. So some organizations are ramping up their investments without necessarily recognizing at first the challenges that they might encounter on the way to achieving real impact with AI in their organization. And sometimes that's like not having or acquiring the data they need to support that application. Sometimes that's not having the right staff to support it. But I think overall, we're sort of going in the right direction, and organizations are kind of going to learn how to get over those hurdles.

Karthik Ramakrishnan [00:32:16]

Yeah, I think you're right. I think specifically a lot of the organizations that were sort of taking a wait and see approach to where this was going or completely ignored it, I think they've all woken up to the possibilities. I think, again, it's a democratization of access to the potential through ChatGPT. I mean, that's just, if you can experience it yourself, then you can extrapolate and say, ah, this is what people are talking about all this time. I get it now. And of course there's some movement, right? We're seeing that every day, like industries that have been, I mean, dare I say, a little bit sleepy or there were not the right incentive structures in place for them to go and do something anyways. But, yeah, I think everyone's thinking about it, and then they're moving forward really quickly. Okay.

There's an amazing career path. What lessons would you say, would you like to share that you think would be valuable for others looking to enter the field, particularly, you know, researchers. Right. Who don't necessarily have a business background or an engineering background. But to your point, I think you married this really well, and I see a lot of researchers struggle with that, which is they're very good deep in the research side, but then the productization side completely escapes them. Right. And they have a hard time bridging that gap. But you've done that successfully.

What lessons would you like to share for someone trying to enter into the commercial side from the research side?

Julianna Ianni [00:34:49]

Interesting. Yeah, I feel like that's a tough one for me because I've not had the business background from the get go, obviously, but I. But I did always have this kind of drive to have an impact and for what I was working on to get to those patients. But I guess I would say for those that maybe don't see themselves as, like, business people or don't see the need for the business aspects of this, realizing the possibilities of the impact that you might be able to have if you put some care into those aspects of the work you do. One of the things that I struggled with a lot previously was just communication. I grew up being extremely shy, and it has taken me decades to work my way out of that, and that was really holding me back a lot at many points in my career. And it's just something that I had to work at quite a bit to get better at. And that was mostly practice and trial by fire.

But that's, I think, been the biggest barrier for me and the thing that propelled me forward the most when I was able to get past it.

Karthik Ramakrishnan [00:36:30]

Yeah, look at you now. I wouldn't have suspected that you were shy in any way, but there you go. That's a job well done. Lots of practice, I'm sure, but no well done. Congratulations to you. Now, look, I think, you know, my many interactions, I think research is generally thoughtful. Right. They're not introverted, but they're just very thoughtful about how they approach things and always want to make sure that they're not, they don't like to get involved in things.

I mean, at least some. Some of the folks I know in things that they don't know. Right. And it's not that they don't care about it. It's just, hey, listen, like, I'm strong here, and this is where I'm going to focus. But to your point, maybe if I can extrapolate your. Your advice is really stretch yourself, because if you want to have the impact, if you want your work to have the impact that you want it to have, then you do need to at least understand and make the effort to understand the commercial side, because that's how really the work that you do gets into the world through that context. So I think if that's fair to summarize or extrapolate that way, I think that's a very, very good takeaway.

Julianna Ianni [00:37:40]

Yeah, I think that's a fair summary. And I would add to that that, you know, this sort of commercial aspect, this is not something that everybody has to do. It doesn't all have to be business. Academia is a great way to make a big impact, but you still have to have a lot of the same skills to be successful in that area, and it's always good to build those skills and sort of sell the work that you're doing.

Karthik Ramakrishnan [00:38:10]

I know you're not saying that you're the representative of all women in the AI industry, but what advice would you give to women looking to thrive in this industry, which even today is quite male dominated from a research standpoint, and it can be intimidating. Some of your friends, very, very smart, and they hold their own, so it's not an issue. But generally, as young women are thinking of entering the industry, what advice would you give them?

Julianna Ianni [00:38:42]

Yeah, I could be here all day. Yeah. And it's tough. I think it's like maybe one in eight in this industry are women. It's crazy low. And my advice is probably the same for everyone, but just to women and other minorities, you have to know that it's a tough place to be and it's going to be harder for you, and it's very unfortunate, but it is just, you can't be hard on yourself about it. I'd say as far as advice, ask as many questions as you can. Don't be afraid to look stupid.

I think there really is no such thing as a stupid question, and you got to just lose your fear about, about looking so stupid. I lost that long ago, but it was there. And I think you alluded to, you know, sort of the imposter syndrome. I think definitely many, many of us have this and we have to acknowledge it and try your best to get rid of it because you deserve to be where you are. You got there because you earned it.

Karthik Ramakrishnan [00:39:57]

Perfect. That's a great ending. Thank you so much, Julianna. I think it was a real pleasure to have you on and awesome takeaways, I think, and would love to have you back as we see more advances on the medical side to just discuss outside of Proscia, even all the things that you're seeing that I'm excited about. So thank you for joining us today. That's all for today's episode of the People and AI podcast. A big thank you to Julianna Ianni for joining us and sharing her valuable insights. If you enjoyed this episode, please subscribe and leave a review. We'll be back soon with more fascinating discussions on the transformative power of AI.

Until then, take care.