We could not be more excited to bring you the next High Tech + High Touch Healthcare discussion with the Jewish Healthcare Foundation to explore how Digital Twin technology is being used in clinical trials. We welcome Prith Banerjee, Chief Technology Officer at ANSYS and Charles Fisher, Founder and CEO of Unlearn.AI to discuss how Digital Twins, Artificial Intelligence and Machine Learning simulate biology and improve the lives of patients. Learn from these world-renowned experts how Digital Twin technology is making clinical trials faster, more accurate and more cost-efficient.
Transcription:
So good afternoon, everyone. This is Audrey Russo, President and CEO, the Pittsburgh Technology Council. Today is a special episode business as usual, and there will be a companion podcast coming soon. So this is titled high tech and high touch healthcare. And it is deep appreciation to the Jewish health care Foundation and the work that they have been doing called lift off PGA, we will actually put a link in the chat section so you can actually see what they're doing over at the Jewish health care foundation. They've been terrific partners with us over this entire 1516 month period, as we really explored so many different things in terms of healthcare and the intersection of tech and the world to come and some visions about the world to come as well. So today is definitely no different in a moment, I will introduce our guests, I'm very excited to have the both of them. And I want to make sure everyone knows that we've muted your microphones, but we have plenty of opportunity for you to ask questions and get get involved in the dialogue. So we I also want to tell everyone, we have Jonathan Kersting. He's here each and every day. And he is vice president of all things media marketing, storytelling. If there's a story to be told, Jonathan will find it and we will highlight it. And we're pretty proud about what lies right at our footsteps all across Southwestern Pennsylvania. Today, we have two guests. And the first is prayeth bannerjee. And he is the Chief Technology Officer of ANSYS. He has a remarkable background and you can look on LinkedIn, you can see the journey that led up to his role at ANSYS. And if you've been following answers, you know that they've been very, very busy during this period of the last 1516 months. So we're very excited to have him. He's also on the board of the Pittsburg Technology Council, the other individuals Charles Fisher, and he is joining us today from where he is, I think living or at least visiting and Lake Tahoe. I asked her view but he said he doesn't really have a view. And he is the founder and CEO of unlearn.ai. And we'll get a chance to to hear their stories. And we're going to talk today about the intersection of of health technology, digital twinning, etc. Lots of stuff. So we're going to pack it in. And I'm going to be quiet now because I'm just going to bring to the forefront. Charles Fisher and press. Right. So hi, Charles. Welcome. Hi, Chris. Welcome. Thank you both for being with us today. I'm going to start quickly with Charles and we're going to say, let's just talk about your background really quickly, your journey to becoming founder of a startup, you know, it took you have a very strong research background. And I think that there's reference to you even studying butterflies in Borneo, to then leading to groundbreaking work at Pfizer. So and then working at a VR startup in San Francisco. I can't keep track of all that. And I know Perth has an amazing backgrounds as well. So can you just talk Can you pack that in a little bit and just tell everyone what your journey has been?
Yeah, absolutely on Thank you for having me. So I call myself a reluctant entrepreneur, and that I never set out to get into business at all. I'm a theoretical physicist by training and I had every intention of being a professor. I did my PhD at Harvard, you can tell I wanted to be a professor because I did multiple postdoctoral fellowships after grad school. And, but at some point, I realized I didn't, I didn't really know what went on in this, this area that academics call industry, they just just industry. I didn't know what that was. And I thought I would, I would try to get some new experiences. So I moved. Actually, at the time, I was living in France, working at a university in Paris, and I moved back to Boston to work at Pfizer as a machine learning scientist, and it bounced around from Pfizer to this virtual reality company in in San Francisco, before eventually deciding to start on learn about four years ago. And by and large, most of my research 95% of my research has been on applications of machine learning and artificial intelligence, the problems in biology broadly, with the virtual reality company being the one the one really different outlier that wasn't related to biology or medicine.
It's great. Well, thank you so much for joining us and I think I know this might be early in the morning for For you where you are, so deeply appreciate that as well. So I'm going to jump to prep. And I think, you know, you're going to be equally as blown away by this background. I, you know, I would like him to just talk a little bit because he sort of like touches every area of industry and entrepreneurship and investment. And I will not do him a good service by trying to capture it. So Chris, thank you and tell tell everyone a little bit about your journey.
Thank you so much, Adrienne. It's absolute pleasure and honor to be here today. So as Audrey mentioned, I have been fortunate in my career to have worked in the sort of three phases. So I've had three phases in my career career. One job was me, are you what you You said you wanted to be in academia, I was successful in getting into academia, I have a PhD in electrical engineering from the University of Illinois, Urbana Champaign. And I joined Urbana as a professor and move the ranks spent 12 wonderful years at Illinois as a professor and then ultimately became the founder, founding director of the computational science and engineering program at Illinois, then went to Northwestern was the department head for for a few years, for eight years, then was the Dean of Engineering at Illinois in Chicago. So hardcore academic for about 20 years doing research on parallel computing parallel algorithms for electronic design automation. During those years in academia, I the startup bug caught me as you know, many professors, depression startups, so I, I started two companies, one was XL cheap, and it was the northwestern second one was buying a cheap and it was at Illinois. And I learned a lot through the processes in how do you raise funding series eight, 5 million Series B, whatever, etc, ultimately sold about the companies again, very small exits. Excited to to Xilinx and other one to get to medium. Then I did sort of phase three in my career, which is what? Joining large corporations, so my first job was heading up HP Labs worldwide. And leading source of research for a large 100 billion dollar company, the long term r&d and so on. I did that for about five years, then, was fortunate to take on CTO roles at two large industrial companies ABB I relocated to, to Zurich, in Switzerland, and then I was CTO at Schneider Electric. I was in France for a while. And then ultimately, about three years ago joined ANSYS as CTO. So I have had the pleasure of working in innovation in an academic setting, in a entrepreneurship setting your startup as well as, as in large companies. So I feel absolutely thrilled to be here today. And CES today is a company which we do modeling and simulation tools, we help our customers in their innovation journey, we help them build innovative products in different areas such as automotive, aerospace, and defense, high tech manufacturing. And we are just beginning to, to look at the area of healthcare and I'm personally as CTO driving the healthcare initiative at ANSYS. We are a company of about 1.7 billion in revenue 5000 employees, headquartered in Pittsburgh, and we're delighted to be part of the Pittsburgh Technology Council.
Great, so everyone can see we have we have two mega humans here that have worked on really hard problems for a long time in their life. And they've touched almost every domain of everything in tech. I mean, I'm probably can't even think of a category that you haven't touched. So I'm very excited to actually focus on healthcare. I think it's been exacerbated during during COVID. And I think like all things, we're rapidly trying to solve some problems that perhaps never would have seen the light had we not had what, you know, what is going on right now in terms of this pandemic, and eventually an epidemic. So let's, I'm going to go back to Perth real quick. And I want to talk about ANSYS a little bit, because they have a deep history in digital twins. I mean, that language might not have been used when they when they started even as am soft, and then you know, moved to become ANSYS. Now, we've talked about digital twins in the past, Chris, you've been with us how and how fast it's growing. But can you level set that level set what digital twins really means and why that's been ANSYS has really been like at the forefront of this right from the beginning.
Sure. So. So as you let me start start from the very beginning, right? So the world around us is governed by the laws of physics, right? And what ANSYS does is we take the physics and we simulate the physics with numerical methods. So we started 50 years ago with with ANSYS john Swanson date with and he had this thing. Let's take a structural problem. And if he applied this much force to it Here's how much it will bend and at this point it will crack, right. So the laws of structural physics are governed by second order partial differential equations called Euler equations. And john Swanson used finite element analysis to solve those things. Then what ANSYS DNA said, Okay, what other physics is there in the world, it's fluids physics, computational fluid dynamics, solve sort of model by navier stokes equations. And so we ANSYS went and bought a company, the leading company in the area called fluent. And with that, we now have the fluid simulation capability with finite volume methods. Then, in 2010, we acquired a company answer and other bits of our company, which gives us the ability to solve electromagnetic equations right now via Maxwell's equations against second order partial. So over the years, we have acquired different physics capability, we have got photonics, simulation, capability, semiconductor ability, so individual physics, sort of fluid structures, electromagnetics, photonics, optics, and the like. And so that's what we have a, you know, Arsenal today, right? The physics that we do that our customers initially wanted to do single physics, like only the structural, but then increasingly start people started working on multiple phases in terms of fluid structure, interactions, fluid structure, electromagnetic interactions, and these things require a huge amount of computing power. And so therefore, we are solvers run on HPC, high performance computing, on premise and on the cloud. Now, the digital twin thing is a relatively new term that has been coined, right, but basically, what a digital twin is, it is a virtual model of an asset, right? Whatever asset the asset can be a transformer, a medical device are a human right, you can have a digital twin. So essentially, there is in this area of digital twins, people talked about a data analytics, you put some sensors on an asset, you collect data, and you look at the normal behavior of the asset, a washer and dryer or whatever, right. And before that asset starts failing, it gives signals it's about to fail. And that's how the whole area of digital twins were working. What ANSYS has done is to combine the data analytics based digital twin with the simulation, because we at the heart know how these assets are working, we can go back to absolute basics of fluids, of structures of electromagnetics. And we can predict exactly how that system will fail. When it will sail, we can create virtual sensors. And so we have this big initiative and emphasis on on the twin builder which team is leading. And that's sort of how we have we have we're solving the world of digital twins, predicting and predictive analytics and preventive maintenance and improving sort of optimal sort of operations in different verticals like automotive and aerospace and manufacturing. And
thank you. Thank you. That was wonderful, great example. So Charles, and 20, and 20, you know, in 2020, and now into 2021, you know, the average American has, has gained tremendous insights into the world's processes for clinical trials. I mean, it's sort of funny that the clinical trials and everyone hasn't has, you know, their own point of reference, it's been common language, we've learned that they're expensive. They sometimes involve recruiting 10s of 1000s of participants. And their failure rates can often be high unlearn AI, if you're pioneering a concept that will help dress many of these issues are hopefully all of these issues. So talk to us about that work. And in developing digital twins for clinical trials.
Yeah, absolutely. So interesting. Lee enough, everyone thought that these files this year, were, we were all waiting for them to read out. They're the fastest things that have ever been run. Normally, clinical trials, and meta medical research takes much, much longer than then did for these these vaccines and these other treatments for COVID-19. So So basically, what you're doing in a clinical trial is you're running, you're looking for a comparison, you have some brand new treatments, and you're going to compare it to the existing treatments, which could be nothing for if there's no treatment, right? So you're comparing it, and you're asking, Is this new treatment better, in some way than what currently exists? And so what we do when we talk about digital twins for patients, is that we create computer simulations of how people respond to existing treatments. So if I'm going to do this comparison in a clinical trial, we don't know enough about biology. So we hear about simulation, physics based simulations of things. But we know very little about biology to be able to simulate it in a sort of bottom up approach where we say this gene does this and it interacts with this. Another thing And so rather than that we build these digital twins with artificial intelligence, so we collect data from, you know, 10s of 1000s of people who, and we observe how they respond to these existing treatments over time. And then we use these these approaches from artificial intelligence that have been developed in the last few years to create computers in a computer simulation engine of people with that disease. So an example is like we work right now and in some trials in Alzheimer's disease. And so if you have a patient comes in and you collect information about their cognitive, sort of their memory status, you have blood tests, you have genetic data, you have brain imaging data, you put that into this simulation engine, then I think you can predict how that patient's disease will progress. So then we can leverage these digital twins of these individual patients within clinical trials, to enable clinical trial sponsors to pharmaceutical companies to run their trials with many fewer patients, so that they can generate the evidence that's necessary to know if their new drug is safe and effective as quickly and efficiently as possible.
Wow. I mean, it's pretty incredible. So why is machine learning such a great fit for tackling the issue and then in the medical trials,
machine learning is sort of fit very broadly, right? tool toolset for creating prediction, making predictions or creating simulations. In our case, the area of generative machine learning generative models, about things which we don't understand, that are too complicated for us to understand how they work. Yeah, so if you look in the past that the way people would build a lot of predictive models, is like some scientists would come up with hypotheses that say, well, this this variable is really important. And that variables important to maybe we'll try to, in machine learning, rather than having sign individual scientists try to understand the details of the process, because it made us process too complicated for us, let's just collect a huge amount of data, and then feed it into special certain kinds of algorithms that are good at making use of those data and discovering those kinds of important characteristics themselves. Instead, we don't really understand biology very well at all. Machine learning, but we have a huge amount of data. Every time someone goes every time you go to the doctor, you're it's sort of generating data that could be used for these things. So there's a huge amount of data, but there's very little understanding. So that's sort of like a perfect interaction of application of machine learning.
So So press ANSYS, you know, as well known, you know, for the work in industrial design, you've set the stage for that in your opening comments, but they also have amazing capabilities. And you you hinted a little bit about the health skin out slash healthcare space. Can you talk about the applications that your champion, so that and how it might affect health care? And what the vision is for that? Oh, he's on you. Hang on. Well, that's not good. asked to unmute. Yeah. Okay, here we go. Sorry. Don't unmute yourself again. Yeah.
I mean, again, what we do, as I said, we look at the world around us governed by the laws of physics. So so let me now talk about the heart, the human heart is the most complicated organ that we have, right? So what we have taken is taking our structural simulation model and the fluid simulation model electromagnetics model, like the heart beats 70 times a minute, right? So we take out the electrophysiology of a heart, I said, depending on the electrical signals that come in, the muscles contract in the heart, because the muskogean model exactly how the muscles contract then because the muscles contract, the the ventricle sort of contracts, and so there's blood that flows from the ventricle to the aorta, right? So we can use our ANSYS, mechanical, our hfss, and fluids, right, and LS, Dinah, all these software to model the heart and we have done a very, very accurate model of the heart. Now, let's imagine that you have a heart disease, right? You have arrhythmia. So you can have two choices you can have, you can take a pacemaker. And so we can now model, a pacemaker from Medtronic and how it is interacting with that failed heart or you can take a medication from Pfizer to address my enemy right. And so essentially Not now, that drug from from Pfizer right needs to be modeled with computational chemistry. So we are partnering with the likes of Schrodinger and ccwg, computational chemistry, right to model those molecules in the Pfizer drug and trying to see how that drug drug Discovery is happening right as Pfizer is when many new drugs and so on, and then how that drug will be, will work on the human organ, right? So that drug is injected. And as it gets injected, we do what is called pharmaco, kinetic for PK PD analysis. So we can do the whole sort of simulation of that drug, the computational chemistry with the PK PD with the virtual sort of simulation of the heart. And then the real challenge in this whole clinical trial thing is I you have to test it on humans, right. I mean, you have we have all seen what happened with the COVID vaccine, right? I mean, the invention the drug happened in one month, it took one year to get the FDA trial. Why? Because Firstly, we tried on on mice, then on on rats, then on to humans, then on 10, humans, then on 1000. Humans, ultimately, you get the thing, right. So what if you could do in silico trials, that is you can create a model of a human eye, it has to be an accurate model, right. So essentially, you have a fat human, a thin human, male, female, children, all kinds of things can be actually simulated. And that's sort of what we are pushing for through in silico trials. But the trouble is, the FDA still is, is relying on sort of, in some victory, the actual human trials. And so we are working with the FDA to say, hey, believe in our computational simulation capabilities, we are so accurate. And thereby if you can do millions of sort of trials, right, you can completely accelerate the way drugs will be put in or your medical devices will be approved by the FDA. So and we cannot do it purely through physics, we understand that AI ml, right, what Charles is working on. It is complimentary. This is why I talked about the digital twin, it's simulation and AI. So we are actually partnering with various companies to take our simulation based approach with the AI approaches to accelerate clinical trials. Interesting.
So then if you flip over to Charles and both of you probably have, I guess, Charles with his background, this whole application for personalized medicine, doesn't it take us down this path towards personal personalized medicine? I mean, he used the example of the heart. But I would imagine the applications just are endless, you know, in terms of cancer treatments. And, you know, I guess it's anything. So do you see the potential chosen new technology to help clinicians? In terms of really, I mean, you talked about treatment options, but what about like, let personalized piece?
apps? Absolutely, yeah, I mean, it's definitely the case that these approaches, you could imagine a future where you go to the physician, and they pull up your digital claim, they have a computer model of you. And they could ask, you know, various questions to see, you know, potentially simulate how you would respond under under different treatment options. It's a future that I think is going to happen. It's not going to happen fast. It is gonna happen very slowly, like not within the next 10 1520 years. Because Oh, yeah, both because the truth is that that simulations of people, like are really actually been not that not that accurate. They're right, they're not precise. And we don't even under like, you look at all simas disease, we have no idea really what causes this hypotheses. But you're talking about genetics, you talk about all of these things, we're not to the point where we can where we can make predictions that are extremely accurate, right. And that is this depends a little bit like we can simulate the the breaking of a bone or something like that. But when it gets down to that the biochemistry and the genetics and the molecular biology of humans, we don't understand that much. And so the idea that, you know, these are going to become very, very accurate simulations that will be precisely tell doctors what's going to happen. It's probably just still a ways away. And then there's all of this other stuff about who pays for it, your insurance got to pay for it. Now. You don't pay for it out of pocket, who's paying for it? Do the doctors want it? How do the doctors interact with it, even though we're new electronic medical records in this country, right. And if doctors hate their electronic medical record systems right now, just even just the basic medical records, there's their user interface problems and basic things. Now you're going to put this simulation on them. I think that this is something where it's going to happen and these tools are going to be used and we're going to see that being used in really specialized cases, right? Even right now you're seeing some things and in say dermatology where you can have AI based systems that can diagnose can help to diagnostic indicators through seeing them specialized. But I think broad based adoption of these things by the medical communities is going to happen. But it's this is going to be a multiple decades journey to that not not just a couple years,
well, you know, I'm gonna jump back to press. But if I can just insert, I think we there are many people who have been in tech and healthcare over these last 16 months that have been surprised at how fast we made transformations during this period of time. So I'm a little bullish and excited that I think there's going to be opportunities that hopefully, Charles will make you happy that it happened within our lifetime. But I actually do, because I think there is an imperative and a whole bunch of learning. And if you pair it with what's happening in digital twinning and, and hopefully ethics, that we begin to, to see that, but I appreciate your pragmatism, and you're certainly more worldly and knowledgeable than I am in the space. But I'm excited. I'm sort of excited on the optimism. So Chris, let's let's, if you want to add something Yeah.
Yeah, no, no. So let me actually address which also, I think he's right there, simulation has its limitations and so on. But what we are trying to do in the area of personalized medicine is a falling, right. So let's say again, that same arrhythmia example that I can't talk about, right. And so there is a generic pacemaker that Medtronic can make and is a generic heart that you can have, right? So Dusseau systems has a has a living heart model, we have a heart model. And so right below the generic heart models, but then when prick goes to a heart specialist, right, this Stickney inside your see a CT scanner or an MRI machine, they can take an MRI of my heart, right? slight layer, by layer and so on is from bits heart, you can take those, there are segmentation tools that will do the meeting. And then that thing comes up with a really accurate model of Pixar, this is how. And so based on it, I can do some simulations, I can say okay, that you are hired to do this, right. But then the technology that exists, right with all the Fitbit sensor, all the sensors that we have on our body, right? You can say my heartbeat at this point will be this and the Fitbit, I actually measure what is happening. So there are tools in the flow, we are working with companies that NSV and cases that tape those CT scans, convert into machine to our songs and then tight with the actual measurements of the whatever you did with your, with your, the partitions, and so right. And using that you can correct your simulations, right. So the whole thing of a digital twin is you have a model, but then you are predicting certain stuff, and you actually get actual measurements, and you correct your model. This is sort of the future you're working towards. And I am quite bullish like that it will not get long, right? Not for all diseases, I think Jonathan's ranking for certain things. where we are headed is we are trying to enable third party applications like clinical apps. So for example, we are working with a company called optimize for AI surgeon. So you are at your laser surgeon, you're trying to get a laser eye surgery, right? We are using ANSYS solvers in the backend to say exactly how you would cut do the laser insert, right. And so we are any going to enable 1000s of clinical apps to be created. Think of it as like iOS or Android app, right? That will use ANSYS solvers in the backend on the cloud running on millions of processors. But a business model will be pay per use, right? That doctor will charge $500. Right. And that's they'll get paid for it. And essentially, as this will get maybe 25% of that. So I mean, this is a very, very exciting future. And again, as I said, you can probably see the excitement device, right? I mean, this is the vision we are exploring enhances the new healthcare vertical.
Yeah, I'm definitely excited by it. And I, you know, I appreciate the pragmatism and the methodologies that have to go into it. I think there's gonna be a lot on ethics. I think there's gonna be a lot of like Charles said, Who's gonna fund it? Who's gonna pay for it? How does it get supported? Who gets access to it? So So, as we wrap up, I want to go to Charles, tell us about your team. Are you ready to open an outpost of unlearned AI in Pittsburgh? Yep.
Well, MC sorba, may or may not because I do think, yeah, we've talked about some transformations that happen in medicine, because there's been a transformation in industry in vain. 19, right. I mean, certainly what's at our company, we are based in San Francisco, and we used to only hire people in San Francisco. That's not true anymore. Right now, we are hiring people all over the country and we're remote workforce. And I think that it's gonna largely stay that way. And I think that this is a really great opportunity is for a lot of San Francisco is very expensive. So it's like it's like, but this opens up a talent pool, right? You're not just sourcing talent from one part of the country, but from the whole country, maybe the whole world. And I think that that's really, really great for industry. So you know, where we're growing. Now, you know, we're focused on, you know, these really complex chronic diseases, Alzheimer's disease, neurological diseases, autoimmune diseases, diseases, where we actually again, don't we don't understand the mechanism of the of these diseases. And, yeah, we're hiring people, people all over the country, in machine learning in stead, biosthetique, we do a lot of work on Biostatistics and how to make sure that we're getting reliable evidence out of these clinical trials. And so, so yeah, we're, if people in Pittsburgh are interested, you know, we are definitely, definitely looking for new, new, talented technical people all over the country.
Okay, so here's the deal. I'm gonna put a little shingle out, right, and in one of our shared working desk, and we're gonna call it unlearn.ai. And I'll talk to you about that offline. So friend, me put a plug for
Pittsburgh, Charles, I mean, I mean, if you look at talent, it is tied to the talent coming of universities, Carnegie Mellon University, I mean, I don't have to tell you is number one in the world. And so any company that is trying to do stuff around AI, right, which is why the Googles and the Ubers, all of them have done outpost in Pittsburgh. And again, so I don't have to tell you, but you know, CMU is there and you paid is another fantastic university to great universities in Pittsburgh, that we are dancers have tapped into. Thank you.
Thank you, Brett for doing that. I think we have a task. We follow up with Charles after that. Yeah.
I am a co one of my co founders, Aaron. He's, he's a Penn State. Penn State grads in new Penn. So we have we have connections, not Pittsburgh, but close by.
We love Penn State, but it's not Pittsburgh. So I mean, I think one of the things we're gonna wrap up, and I know, we could have spent more time with the both of you, you're both lovely, warm and brilliant. The both of you are. So this topic is is really important. And I know we could have lots and lots of this. I don't know if we're going to do a podcast, in addition to this, but we really do want to do a deep a deep dive. So I so press, Are you hiring right now? Are you doing any hiring,
you're hiring, and we are hiring across the company in all areas. But I as CTO, I'm responsible for a few initiatives, I'm responsible for our AI ml initiative, across simulation. So I kind of know what I mean, anyway. So that's where our AI talent designing. My healthcare is part of my team, that digital twin is part of my team. And all these teams are actually hiring.
It's great. It's totally great. So we'll put links out there, I want to thank both of you, we didn't get to everything. I know that there's a digital twin consortium that ANSYS has rolling, they've been one of the seven founding members. And I think there's a link, we will probably put a link out there so that if people want to know more about it, but ANSYS really has been part of the magic of Pittsburgh, and they have deep roots, in terms of changing the way people work in the way people model. It's, there's probably no other company like them in probably the world. And they've just been just incredible. So Charles, it's, it's wonderful to meet you. And we were gonna make some connections with you after this. And I'm going to share some of the things that are happening here in Pittsburgh, because we know it's true. It's not just that we're proud, we know it's true. And I applaud you for the work that you've been doing. Because it's critical. I say toast to the future. I think the future is closer than we think. And I just appreciate you both taking the time with us for sure. So thank you both. And if you want to know more about their companies, easy to find easy to gain access to printers very accessible here in our region. And I'm sure Charles won't be shy about connecting with the people that will connect him with so thank you both Jonathan, what's what's up anything up?
Tomorrow is a can't miss business as usual because we had FedEx stopping by talking about their $2 billion carbon neutral initiative that they want to hit by by 2040. Lots of business opportunities they've got great. We're up to so should be a lot of fun and very
well, Charles, thank you again. Thanks for being with us. Thanks to the Jewish health care foundation for connecting us all and supporting us. Thank you prefer that Everyone stay safe. And hopefully we'll see you all soon. Thank you. Bye bye. Thank you
Transcribed by https://otter.ai