Jon Krohn: 00:00
This is episode number 579 with Dr. Wardah Inam, co-founder and CEO of Overjet. Today’s episode is kindly brought to you by Z by HP, the workstations for data science.
Jon Krohn: 00:16
Welcome to the SuperDataScience podcast, the most listened to podcast in the data science industry. Each week, we bring you inspiring people and ideas to help you build a successful career in data science. I’m your host, Jon Krohn. Thanks for joining me today, and now let’s make the complex simple.
Jon Krohn: 00:47
Welcome back to the SuperDataScience podcast. Today’s guest is the brilliant engineer, computer scientist and AI entrepreneur, Dr. Wardah Inam. Wardah is co-founder and CEO of Overjet, an AI startup that has raised 79 million in venture capital to transform dentistry with machine learning, enabling dental care to become more precise, more efficient and more patient-centric. Previously, she co-founded uLink Technologies, a startup behind AI-driven power grid systems. She has also served as lead product manager at Q bio, a VC-backed AI startup in the healthcare space and is a post-doctoral researcher at MIT’s renowned CSAIL, the computer science and AI lab. She holds a PhD also from MIT in electrical engineering and computer science.
Jon Krohn: 01:34
Today’s episode focuses more on practical applications of machine learning and growing an AI company than getting into the nitty-gritty of machine learning models themselves, so it should be broadly appealing to both technically-oriented and business-oriented folks. In the episode, Wardah details how Overjet not only classifies images, but quantifies dental diagnoses with computer vision and enabling models to answer questions like how large is this cavity? She also details how natural language processing can be essential for determining the correct dental diagnosis. She talks about the data labeling challenges firms like Overjet need to overcome in order to enable machine learning models to learn from noisy real-world data. She provides her tips for building a successful AI business, and she fills us in on what she looks for in the data scientists and software engineers that she hires. All right, you ready for this jaw-dropping episode? Let’s go.
Jon Krohn: 02:34
Wardah, welcome to the SuperDataScience podcast. I’m so excited to have you on. Where in the world are you calling in from?
Wardah Inam: 02:40
So, I’m in Los Altos Hills in California.
Jon Krohn: 02:44
Nice. Is it beautiful this time of year over there in the springtime?
Wardah Inam: 02:48
Pretty much, yeah. It’s so beautiful most times of the year.
Jon Krohn: 02:53
Oh, lucky. I don’t know. I could be working anywhere now since the pandemic, we’ve decided to not go back to offices. We have an office, but we only use it occasionally. For some reason, I’m just still stuck in New York. I put up artificial walls around why I can’t go somewhere beautiful like California, or like Colorado, or just enjoy the outdoors. I just look at cement all day. Maybe that’s what I like. So we met in New York, actually. So we met at the ScaleUp:AI conference, which was held by Insight Partners in New York and it was walking distance for me. So, I live downtown and I was able to walk over to that conference and we met at lunch. So you were getting ready to go on stage and speak and I was just interrupting you while you tried to eat and found you really fascinating and asked you to be on the show and you said yes. And now here you are, thank you.
Wardah Inam: 03:49
Oh, thank you for having me, Jon.
Jon Krohn: 03:52
So you’re the co-founder and CEO of Overjet. Overjet has raised 79 million from the likes of Insight Partners themselves, from Crosslink Capital from General Catalyst. So, it’s a really fast-growing, venture capital backed company using machine learning, using artificial intelligence, to innovate in the dentistry space. So, I understand that the name Overjet itself is an inside term that dentists use, but I don’t know what it means. Can you elaborate on what Overjet means?
Wardah Inam: 04:26
Yeah, so Overjet is a dental term. So basically, your top teeth are in front of your bottom teeth for your bite to close properly and that distance is around two to three millimeters. So, if you don’t have that distance between the top and the bottom, your bite is not going to close properly, and you’re going to have pain. If you have it too much, then you have buck teeth where people might make fun of you. So, it is about the precision of dentistry where you need to have … millimeters matter a lot and even the two to three millimeters matter. So we chose the name, primarily looking at the precision in dentistry and how measurements really matter.
Jon Krohn: 05:07
Cool. That is a really good name choice. Yeah, so Overjet is a clinical intelligence platform, driven by machine learning for precision dentistry, as you say, is the Overjet name belies. So, you’ve mentioned in other interviews that the lack of agreement on treatments by particular dentists makes dentistry more of an art than a science. So, how can machine learning used by companies like Overjet improve this situation and turn dentistry into more of a science?
Wardah Inam: 05:43
Yeah, no, that’s a really good question, Jon. So especially in dentistry, it’s present in other healthcare fields as well, where there’s a lot of variation in diagnosis in general. So, two dentists looking at the same information might disagree with each other on what they’re seeing. A lot of it is because they’re eyeballing distances and measurements, they’re making subjective decisioning and this is where vision really comes into play for us and especially what Overjet does, which is quantify disease. So at Overjet, we actually quantify disease rather than just detecting disease. So, for example, if we’re talking about bone loss, it’s a measurement, that’s a bone level above a certain threshold, and once you can measure it, now you’re talking about, we both can agree anything above, say 2.5 millimeters, it will be considered bone loss. And once we measure it, there’s no subjectivity there. So, what we are doing at Overjet is we are quantifying disease that helps us make very objective decisions, that help us make better decisions for the patients themselves. And that helps move the industry more towards a science and also move towards more evidence-based dentistry as well.
Jon Krohn: 06:53
Super cool, I love that. So, it’s not just classification, as machine learning practitioners, we see lots of examples, even in early machine vision problems and data sets that we work with. We’re often working with classification. So, is there a tumor present in this tissue or not? In your case, it could be, is there a cavity present in this image or not? And so Overjet goes beyond that. It’s not just image classification, it’s quantification, it’s saying, “Yeah, how many millimeters of bone loss do we have?”
Wardah Inam: 07:28
How large is the cavity. How does that interact with the rest of the tooth, for example, similar to … or you do detection as well, So you’ll say, okay, is there a cavity in this particular area. That’s one, but then we go a step further and we say how large it is, how it relates to the anatomy, how much of that crown area of the tooth has been decayed. So we can actually get to a very accurate decisioning based on the measurements.
Jon Krohn: 07:54
That’s super cool. Sounds like it would be a tricky machine learning problem. I guess we’ll get into that later in the episode, particularly around, getting sizes accurate in images, that kind of thing that comes to mind to me as something that would be tricky when we’ve labeled the data. So, as a specific kind of example, of a specific kind of dental disease and how Overjet can make an impact, I’m aware that every year there are three million cases of periodontal disease in the US alone. So, that’s about 1% of people in the US are diagnosed with periodontal disease. In an interview last year, you mentioned that periodontal disease is likely heavily underdiagnosed. And, so what are the drivers for this underdiagnosis and what role could data and machine learning play in identifying those currently undiagnosed cases?
Wardah Inam: 08:55
Yeah, no, the area of periodontal disease, according to the CDC, it’s more than 40% of the adult population above 30 years has periodontal disease. So, we think that’s a little more than what’s actually what we see in the data. So, I think probably CDC needs to update these metrics, but on one side, you’re saying it’s 1%, what we are seeing in the practices is around 5% to 7%. But the population that we’re seeing that has the disease that should be treated is much, much higher. So, it is one of probably the most underdiagnosed diseases in the US particularly, but also in the rest of the world. And why is it so important is that periodontal disease is also linked to diabetes. It’s also linked to cardiovascular health, so it’s basically an infection in your mouth that leads to inflammatory response and it causes other immune system to go haywire as well across your whole body.
Wardah Inam: 10:03
So, it is something that is very, very important and it has been well studied, why it’s so important to actually not only diagnose, but also treat, but it is going underdiagnosed. Here where Overjet comes into play and computer vision helps is first is, we need to be detecting the disease. I started the conversation with bone loss, for example, and that’s one of the most important metrics when it comes to me measuring predominant disease. It’s after the infection has been there for a bit, your start to have bone loss that could lead to tooth loss and other issues as well. And when you’re eyeballing this, even if the tooth is slightly aligned differently, one person might say there’s bone loss. The other person might say, there’s no bone loss.
Wardah Inam: 10:56
In this diagnosis, there’s also the hygienist involved and the dentist, there are two parties involved. And they’re also taking hundreds of other measurements when they’re making this diagnosis too. So, every practice we’ve entered into, yes, the practice does under-diagnose the disease and where Overjet helps is we identify for every patient, every measurement that is needed to make the diagnosis. And if we find some of the data which is present and other might not be, either has it not collected, we can flag them to collect that. If all the information is there and the treatment has not been identified, we will flag that for the practice as well to look at and help determine whether there is predominant disease.
Jon Krohn: 11:41
Whoa. So not only does the Overjet platform just receive information and help with diagnosis based on that information, but it can also provide feedback to the users. So, it can say to a dental practitioner, “Hey, we are missing a piece of information in order to be able to make this diagnosis properly, please obtain that information and provide it to us.”
Wardah Inam: 12:02
Yeah, no. So here, exactly, not only do we identify disease or quantify disease, but we also first help make sure that the right data is present. So, the right protocols are being followed. The right data is present to make the diagnosis. If there’s some indication of disease and the complete data is not present, we will highlight that to the dentist or the hygienist, whoever is in the office to be able to collect the right information so that we can provide a complete diagnosis and treatment planning option as well.
Jon Krohn: 12:35
That is super cool. Do you have any other kind of informative use cases, Wardah, of how machine learning or data science plays a role in the Overjet platform for preventing or identifying some kind of dental condition?
Wardah Inam: 12:53
Yeah, so I think the way … What Overjet does is we look at every condition that can be found on 2D x-rays currently, and we are also working on the other formats in the future.
Jon Krohn: 13:07
Cool.
Wardah Inam: 13:08
But here, one of the important use case or applications for it is actually in the insurance space. So, not only do we work on the practice side to help practices perform better, but we also help insurance companies run their operations more effectively as well. And currently they, for example, process all this information manually. So, you have dental reviewers who look through the data. So if say, go and get a crown, or you need a crown, your dentist is going to submit that information to the insurance company. The insurance company is going to have a dental specialist and then a dentist look at it. The dental specialist will say, “Is this the right information?”
Wardah Inam: 13:45
The dentist will say, “Okay, if this is the right information, is there medical necessity?” And if the answer is no, they’ll not preauthorize the treatment. If the answer is yes, and you have the condition, they will preauthorize it as well. This is all very subjective, it takes a long period of time and it’s very costly and Overjet and AI here solves this and automates this process so that it can happen much faster, much more cheaply as well, as more accurately, such that these are objective decisions being made.
Jon Krohn: 14:22
Super cool. This episode of SuperDataScience is brought to you by Z by HP. Get rapid results from your most demanding data sets, train data models, and create data visualizations with Z data science machines, which come in both laptop and desktop workstation options. The data science stack manager on these Z by HP machines provides convenient access to popular tools and updates them automatically. So this helps you customize your environment easily on either Windows or Ubuntu. Find out more at hp.com/datascience. That’s hp.com/datascience. All right, now, back to our show.
Jon Krohn: 15:04
I wasn’t aware of that application space, but that makes a lot of sense. Being able to not only automate and improve diagnosis in the dental office, but also allow the dental insurance companies to be streamlined as well. Super cool.
Wardah Inam: 15:20
And with that, I think that was our claim to fame actually, when we started off. And right now we’re serving about a majority of the top carriers. We’re presenting about a hundred million members. So, that is half of the US population’s data is going through now Overjet’s platform-
Jon Krohn: 15:35
Wow.
Wardah Inam: 15:36
… in order to make a determination on whether the claim is appropriate or not.
Jon Krohn: 15:41
That explains the amount of money you’ve been able to raise, Wardah. So are there any particular kinds of challenges that you deal with that are particular to the problems that you’re tackling? So, are there things that crop up in your space that are interesting problems that impact machine learning?
Wardah Inam: 16:02
Yeah. So I think, a bunch of [inaudible 00:16:06] … one of that which you mentioned earlier as well, which is that variation of diagnosis and why that’s a problem is not just because we’re helping solve it, but it’s a problem when we actually annotate our data as well. So, if three dentists look at the same condition, one might say a certain thing, another might say another thing. Here, there are different approaches to it. You can try to get a consensus. You can have multiple dentists disagree. Then you can have a … or a radiologist, try to get a consensus, or you could actually use all the labels and feed them into the models and let the models learn what is the right information, where you might have a disconnect there?
Wardah Inam: 16:44
We’ve tried different approaches and interestingly enough, we are actually finding that … which we haven’t seen much in the literature, but we’re finding that just not going through to a consensus, but actually feeding a lot of this information to the models and letting the models learn has got us better results. What we do make sure is, of course, we have to make sure all the labelers are well calibrated and they’re reviewing based on the guidelines that were provided, et cetera, as well.
Wardah Inam: 17:16
Another thing, very interesting that we have to deal with is the quality of the data, as well. As I mentioned, we deal with insurance data and insurance data is … 70% of it is digitized, so that will come in a digital format. 30% of it is basically analog data that has been digitized in somewhere or other, where they’ve actually scanned it, that they might have then printed it out on a paper, they might have folded it. Then they would’ve put a tea or a coffee mug onto that page. Then they would’ve scanned it again and sent it off to the insurance company. So, you can’t be sure what you get at the other side. So, we have to deal with a lot of data which is low quality in terms of the resolution, but also in terms of all these artifacts that might appear like disease, if it was disease and we have to build a lot of very robust models that help identify these conditions, but also crop these things out, rotate them, make sure that they’re oriented correctly, make sure that the DNA is in the right format for us to actually make any kind of diagnosis or quantification that we need to make.
Jon Krohn: 18:33
Cool. Yeah, that does sound like something challenging. So, how do you resolve something like that? How do you deal with a coffee mug stain artifact in an image?
Wardah Inam: 18:46
So sometimes, of course if it’s … a lot of times there might be an x-ray image on a paper and that there might be a coffee mug where there was in the image, so you can actually just crop it out. So we have models which detect where the actual information is and everything else gets cropped out and where we can determine that, or just we have trained with a lot of data as well, to be able to … and a lot of noisy and messy data as well, to be able to learn these different artifacts and that they don’t matter. So, it is a challenging problem and we heavily rely on augmentation, et cetera, too, to make sure that we can create different conditions as well. But it is something very interesting and something we have to deal with all the time.
Jon Krohn: 19:37
Yeah. Data augmentation seems like a great way of overcoming the issues in that kind of scenario. And yeah, it sounds like lots of big challenges and lots of data for data scientists and machine learning engineers at Overjet to be working with. So, there might be some listeners out there who are thinking it would be a pretty cool place to work, and I’ll have more information for them later on in the show. For now, I’d like to stray a little bit away from technical machine learning questions to a general question about Overjet, which is what was your experience as you began providing tools to the dentistry industry? So, were you welcomed with open arms or was there a lot of hesitation to what you’re doing, to automating what has historically been a manual industry?
Wardah Inam: 20:31
Yeah, so previously I was in the medical imaging space, so MRI data. When you are dealing with AI for MRI, radiologists are threatened. So, they’re threatened for … they believe that AI will take their jobs and if you look at it, there’s some truth to it, which is the fact that’s what they do. Radiologists’ main goal is to diagnose on radiographic information. What’s different about dentistry is dentists don’t think of themselves as radiologists, They think of themselves as surgeons. So yes, they diagnose on x-rays. So, they’ll spend some time, maybe a minute or two when they’re with you while you’re in the chair, to look at that information. But most of the time they spend is either chatting with you to explain that information to you, or they’re spending drilling in your mouth, or doing some kind of surgery.
Wardah Inam: 21:30
So the interesting thing is, which is unique about dentistry, that dentists, the way they perceive themselves helps us in terms of, they think of all these things that they can be. If technology can help them do these things, they do accept it with more open arms than you might see in the other healthcare fields. So, that’s been an advantage to us and just in general, in dentistry as well, because you have had less technologists focused on it, they do feel that you need more technology in dentistry and good technology. They’ve had technology, which is, it seems like it’s from the 90s et cetera, and software. And now when they see exciting, cool tools, they’re actually gravitating towards it. So, in terms of timing, I think it couldn’t have been better for us, where at least for now, we haven’t had that kind of resistance that I saw in the medical side. And that helps us, of course, sell more easily into these companies, but also for them to use it so that they can actually improve the lives of people as well.
Jon Krohn: 22:39
That’s awesome. I’m so glad to hear that difference. It hadn’t occurred to me that there would be that difference, but as soon as you explain it, how the dentists have this broader range of functions, than just doing diagnosis of images, that’s actually just a small portion of their job. So, if you can make that part of their job easier, then they can focus more on surgery, or on bedside manner, on dealing with patients and helping them get through what could be a traumatic drilling experience. So, you mentioned there working previously with medical diagnostic imaging with MRIs. So, was that at Q bio?
Wardah Inam: 23:22
Yes.
Jon Krohn: 23:24
Yeah, so that was another startup that you were involved with prior to Overjet and that’s a company focused on simulation and quantification of human physiology. So, how is your experience at Q bio helpful for co-founding your own startup and running your own startup later on?
Wardah Inam: 23:42
Yeah, no, I think it was a very interesting experience with … The startup was started by a second or third time co-founder or founder that meant funding was easier, other processes were easier. But what I learned, I think, from that experience was how important it was to hire great people and how you had to go out of your way to hire those great people. So even, in this case, even though it was founded by this third time founder whose last company was a firm worth, I think billions of dollars, he still … To recruit me, he flew all the way, had a chat with me while I was still right out of school in that sense. But he felt that I was important for the company at that time. So, I think that was my biggest takeaway from the company. I think that has helped us succeed as well, because for me, once we identify who we need to hire, then I’m going to go out of my way to make sure that we hire them and that has led to the success at Overjet as well.
Jon Krohn: 24:47
Cool. That has been a recurring theme recently on the episode, we’ve had a number of guests recently who have said the big driver behind their startup success was hiring the right people. So yeah, it’s great to hear you reinforcing that message here again today. And I love that idea of, especially in this world where we’ve gotten used to now being able to stay home, if the CEO of a company flew out and came to see me in my city, I would instantly think, “Wow, this person really does see me as the future of their company and I should really be seriously considering this opportunity.” That’s great, and that’s really good advice.
Jon Krohn: 25:30
So, in addition to the many patents that you filed at Overjet and no doubt, the innovation that you were involved with at Q bio, you’ve made inventions in other fields as well, Wardah. So as an example, during your PhD at MIT, you worked on autonomous microgrids. So this is a solution to helping people who lack access to electricity. So, today there are about 900 million people on the planet that do not have electricity. So, it sounds like these autonomous microgrids that you’re involved in the development of could play a role in ameliorating that situation. I know you have some experience with this hands-on because you ran field tests in rural India. So, I think you’re the right person to ask, how do microgrids work and how could they solve this electrification problem?
Wardah Inam: 26:25
Yeah, so when we started off on this problem, there were 1.2 billion people who did not have access to electricity. I don’t think we actually got it down to 900 million, but there has been a lot of work in trying to electrify the areas which don’t have access to electricity. Microgrids in general were basically an interconnection of electric systems, which could help provide electricity in a very low cost manner to the areas, where the autonomous piece came about because we were not only identifying what were the loads and the sources, so that means the solar panels or what were the loads such as, lights or fans, et cetera, that were connected onto the system automatically, and then helping control what is called demand response, have them control the demand in such a way that we can best utilize the resources such that we maximize the use of electricity. The cool thing about these systems were that they did not need any planning. So for example, if you look at the grid system in the US or any developed world, it has gone through a lot of planning that’s because it was done back in the day where you needed all the planning to really structure everything. But with the current technology, could you have self-learning grids where you could understand what was happening, who was getting connected, what source was getting connected, and then rearchitect itself to provide the best electricity, at a low cost?
Wardah Inam: 28:00
The interesting thing is this actually not only is a problem in the developing world, but it got used in Australia and other areas as well, because microgrids are more important, especially after you’ve seen the California fires and electricity shortage, et cetera. So you have grids, which can be disconnected from the main grid and then self-function and self-regulate the power there as well. So, at uLink where we were, not only was I doing my PhD, but we were also commercializing this technology and working with both developed world [inaudible 00:28:43] systems as well, but also in the developing world where we helped electrify some villages where we would go in, install these systems and help provide the electricity there and basically provide a proof point on how low electricity can cost if you actually did not have to bill for that much robustness, but you actually built for maximizing the time that the load was available.
Jon Krohn: 29:14
Wow, that sounds like such impactful work. So you were doing that during your PhD. Was this related to your experience of the Harvard Innovation Lab?
Wardah Inam: 29:24
So no, at Harvard Innovation Labs, that’s where Overjet got started. So, after finishing my PhD, I was more focused on clean energy and that’s where the microgrid work came about, but I realized that really wanted to work in healthcare as well. So, after finishing my PhD, I did a postdoc in healthcare at CSAIL, working on how do you use wireless signals to detect breathing heart rate and use machine learning on that, to identify different issues of medical conditions, or other issues that might be identified. So, I made a conscious effort moving from clean energy towards healthcare, and that was the transition I made. And then when Overjet started, we actually got incubated in Harvard Innovation Labs for the first two years.
Jon Krohn: 30:22
I see.
Wardah Inam: 30:22
That was an amazing opportunity.
Jon Krohn: 30:25
I see, I see, I see, I see. So let me try to recap, I’m probably going to get some part of this chronology wrong, but you can correct me where I get it wrong. So, while you were doing your PhD at MIT, part of your research was on these autonomous microgrids and you co-founded uLink Technologies to commercialize those autonomous microgrids while you were doing your PhD?
Wardah Inam: 30:50
Absolutely.
Jon Krohn: 30:50
Got that, okay. And then after you finished your PhD, you did a postdoc at the renowned CSAIL lab at MIT. We’ve had a number of guests on the show from CSAIL, most recently, Professor Tim Kraska in episode number 571. He was talking about collaborative, no-code tools that he had developed for data scientists at CSAIL and was now commercializing. So CSAIL, yeah, one of the best known computer science and AI labs in the world, if not the best known. And so you were there doing a postdoc after your PhD and that allowed you to make a bit of a transition from the electricity work towards healthcare?
Wardah Inam: 31:39
Absolutely.
Jon Krohn: 31:40
Okay. And then from there, from that postdoc, you made the jump into a healthcare startup Q bio, which we talked about. So, that was the company working with medical diagnostic images with MRIs. So, what caused you to make that jump? How did you make this transition from academia into the entrepreneurial space? Was that something that you always knew you were going to do?
Wardah Inam: 32:11
Yeah, no, I really wanted to have an impact, have an impact in my life, and academia is great, but a lot of the things take a long time to manifest in the real world. And for me, the most exciting thing is when something I build actually gets utilized. Because of that, I always was, even while I was doing my PhD, very attracted to entrepreneurship and doing things on the side and trying to commercialize different ideas, et cetera, because I think that’s just been, what’s gotten me very excited. And then there was a pivotal time where I had to decide whether academia or industry, and I gravitated towards trying to have more impact in the near and short-term future.
Jon Krohn: 32:59
That resonates with me. Having done a PhD and then gone into industry, I completely understand how, yeah, things can sometimes, not always, but can often feel like they’re moving a bit slowly in academia. And yeah, it certainly doesn’t happen very often that an academic raises $80 million to be scaling up their dental machine learning startup. So, I can see why you went down the path that you did, and you certainly are making a massive impact Wardah, it’s awesome to see. So then after Q bio, you then got involved in the Harvard Innovation Labs program after that, yeah? And so did you go into that innovation labs program with this dental machine learning idea in mind? Or is that something that came out of the incubator experience, out of the Innovation Lab experience?
Wardah Inam: 33:54
No, so we went into it with the idea in mind, so I basically … The way it started was I went to a dentist and the dentist gave me a treatment plan, which was very different than what I had received before. And that got me interested in dental diagnosis and why there was this variation. I asked for my x-rays and started understanding, reading x-rays 101, and realized that there was a huge variation. You go to 10 different dentists, you’re going to get 10 different opinions, and wanted to solve this. So, I left Q bio knowing that I wanted to work in dentistry, knowing I wanted to help improve the care that was being provided and then I started working on Overjet, and the idea started forming more. And then we applied to Harvard Innovation Labs with the idea. I don’t think it was completely evolved to where it is right now, but at least we knew what we were trying to do.
Jon Krohn: 34:59
Cool, and then, so what’s that like? I guess, let me actually ask the question in two parts. So, it sounds like you keep saying they, prior to applying to the Harvard Innovation Lab. So, you selected co-founders I guess, before you applied to the Harvard Innovation Lab. So, how did you decide on who your co-founders would be?
Wardah Inam: 35:21
So, one of my co-founders we had just worked together at Q bio. So, I knew him, I had worked together with him. He was heading computation, I was heading product side. So, we knew we worked together. He had gone off to work at Amazon after that. So, as the idea started to form more and more, I started connecting with him and we started brainstorming more about it. Then the second person was a dentist, who was of course needed in this process as well, and who was also from MIT and Harvard. So, we had some connections there and that’s how it came about.
Jon Krohn: 36:03
Nice. And then once you got into the Innovation Lab Program, what was it like experiencing that? How was it helpful for getting Overjet rolling?
Wardah Inam: 36:13
Yeah, so I think for us, it was very crucial. First because we did not have funding and we got a space to work at, and this is pre-COVID. So, you didn’t want to work at home as well, and you wanted a place to work. So, we had a place to work, but also there was mentorship around it. I think what I really loved about the Innovation Labs was the fact that they actually, rather than focusing on teaching you entrepreneurship, they were focused on enabling you to do your startup more effectively. So, anything that you needed to make that happen, whether it was some sessions that were happening that you could just attend, or there were VCs coming into the I Lab to think through your ideas, et cetera, so that you were surrounded by other people who were doing the startups as well, who were not just … it wasn’t a school project for them. And to make it better, they were a lot of people around you were helping you support and build your idea further too.
Jon Krohn: 37:13
Nice, nice. That sounds like an amazing curriculum and no doubt had a big impact on your ability to develop and scale Overjet effectively. I also noticed that when you were doing your MIT PhD earlier, you minored in entrepreneurial management, and we had a little bit of a conversation about this before we started recording. So, I thought it might be fun to reflect your thoughts on air. So, while doing that MIT PhD, you minored in entrepreneurial management at the Harvard Business School, which is renowned, as it’s probably the most prestigious business school in the world to be able to get into. So, it sounded to me like that could be a super valuable program for helping you get going with any startup, but it sounds like you might think real-world experience is actually a lot more valuable than learning about entrepreneurship in the classroom.
Wardah Inam: 38:05
Yeah, so I think that this is something that I … because I was always interested in entrepreneurship, and with being in school that long, you actually want to learn as much as you can as well. So, I did take quite a few entrepreneurship courses. What I realized was, at least they were very focused on the early stage, like the idea refinement, et cetera, or just thinking of it on paper. I think, especially with entrepreneurship and starting a company, the real challenges and how you have to overcome them and how you have to still go, even if on paper something doesn’t look right, even then still going forward with it and believing in it is so much more important. And that was something that I felt that lacks in school, but it’s I think, something that is getting improved with things like Harvard Innovation Labs, where would they give you an opportunity to really build out your company forward, that some other schools can also leverage and learn from.
Jon Krohn: 39:16
Nice, I love that answer. So, for people who are listening, who might like to get their hands dirty in an entrepreneurial opportunity and see how it works, if people want to join Overjet as a data scientist, or a machine learning engineer, or maybe some other role, are you doing any hiring? And what do you look for in people that you hire?
Wardah Inam: 39:38
We’re absolutely hiring. I think we have a high bar in terms of people we want to hire, but we have many, many roles open, even if you don’t see it on the website, we are hiring for any and every role in related to data science, machine learning, peer vision, natural language processing, which we didn’t get into, but that’s also some things we do. So, please do apply here. What we really look for is people who not only are good at that particular skill set, but also want to take ownership of the problems. We are a startup, that means that ownership piece is very important to us. And also, especially people who believe in the problem that we’re trying to solve. I think if you have the three, I think you’re a sure shot fit for Overjet. Even if you don’t have, I think, when people join Overjet, they fall in love with the mission that we’re trying to achieve here, because I think it’s something that’s extremely needed for healthcare for everyone, including us, because everybody gets impacted by poor oral care, if that’s the situation.
Jon Krohn: 41:00
Everyone’s got a mouth, everyone’s got teeth. Yeah, so the three things that you look for are the technical skillset, obviously, for whatever the role is, data science, machine learning, engineering, maybe NLP engineering. The second thing is taking ownership of problems. And the third thing is believing in the problem being solved. So, we didn’t talk about natural language processing earlier. I didn’t know about that aspect of what you do, but the show isn’t over. Let’s squeeze it in. How do you get natural language processing involved in dental care?
Wardah Inam: 41:34
So, not only do we actually look at images, that’s one part type of diagnostic data that’s present. Yeah, actually I look at the dentist notes as well.
Jon Krohn: 41:44
Of course.
Wardah Inam: 41:44
These are narratives, short form, not grammatically correct, and all the issues that you have in healthcare notes that we have to deal with. And we process that information not only on the practice side, but also on the insurance side, to make better diagnosis because not all information might be present in an x-ray. So, for example, like a fracture, a fracture does not … you don’t see it in an x-ray, but that might be written in the note. And this is dentist writing these notes, so it does not appear as the right spelling and the way you would see it, it appears in all different ways and you have to deal with that kind of language as well.
Jon Krohn: 42:27
Nice. I’m glad that we squeezed that answer in, but yeah, now we are reaching near the end of the episode. So, regular listeners will know that it is getting close to that time when I ask for a book recommendation. So Wardah, is there anything interesting that you’ve read lately that you can recommend to our listeners?
Wardah Inam: 42:45
I think the book that had some impact on me was Being Mortal by Atul Gawande. I think it’s a deep read, so it’s not a light read and I think you probably can do it over a holiday or something because it really makes you think and really makes you think about your life and especially the last phases of it, where even if you are not going through this, you have family members who have gone through this and people you love, et cetera. So, it is a book that makes you think about life more deeply.
Jon Krohn: 43:20
That sounds like a really good recommendation. And it is something that, yeah, I think on deeply from time to time. Actually, I had many years ago, one of my best friends did a course at university on grief, death, and dying. It was quite off piste relative to the other things that he was studying. And he would not shut up about how valuable taking that course was for his appreciation of life and yeah, being mortal. So, that sounds like a great recommendation. All right, so Wardah, you’re tremendously successful, you are a brilliant speaker. You’re working on amazing AI problems and scaling them up, making a big impact across the world. So, I’m sure there are lots of listeners out there that would love to hear more from you. So, how can they follow you online?
Wardah Inam: 44:14
So thank you for that, Jon. So I am on LinkedIn and Twitter @WardahInam, my first name, my last name. So, those are the two areas that people can connect with. You can also email me, it’s my first name @overjet.ai. So, happy to help.
Jon Krohn: 44:36
Nice, amazing. Thank you so much, Wardah. Thank you so much for being on the program. And maybe we can catch up again in a couple of years to see how Overjet is doing, or maybe how your next venture is taking off. We’d love to hear from you again sometime in the future. Thank you so much for making the time to be with us on the show.
Wardah Inam: 44:56
Thank you, Jon, for having me.
Jon Krohn: 45:03
Wardah is so cool and such a smooth communicator of advanced topics, be they technical machine learning topics, or practical commercial ones. In today’s episode, Wardah filled us in on how thanks to painstaking labeling efforts and image augmentation. Overjet is able to not only classify dental images, but also quantify aspects of them, for example, by identifying how many millimeters of bone loss there are in a given tooth. She also talked about how Overjet’s technology is used not only by dentists, but also by insurance companies, demonstrating how one AI platform can be useful across multiple industries. She talked about how AI solutions can reduce the underdiagnosis of common diseases like periodontal disease, how AI-driven autonomous microgrids can enable for lights to stay on in areas that have been disconnected from the primary grid. How getting involved with startups can enable you to make a bigger impact with innovations than an academic career and how she looks for data scientists and software engineers who exhibit technical knowhow, take ownership of problems, and demonstrate an interest in the problem her company is solving. As always, you can get all the show notes, including the transcript for this episode, the video recording, any materials mentioned on the show, the URLs for Wardah’s social media profiles, as well as my own social media profiles at www.superdatascience.com/579. That’s www.superdatascience.com/579.
Jon Krohn: 46:21
If you enjoyed this episode, I’d greatly appreciate it if you left a review on your favorite podcasting app or on the SuperDataScience YouTube channel. I also encourage you to let me know your thoughts on this episode directly by adding me on LinkedIn or Twitter, and then tagging me in a post about it. Your feedback is invaluable for helping us shape future episodes of the show. Thanks to my colleagues at Nebula for supporting me while I create content like this Super Data Science episode for you. And thanks of course, to Ivana Zibert, Mario Pombo, Serg Masis, Sylvia Ogweng and Kirill Eremenko on the SuperDataScience team for managing, editing, researching, summarizing, and producing another super cool episode for us today. Keep on rocking it out there, folks, and I’m looking forward to enjoying another round of the SuperDataScience podcast with you very soon.
