Innovating AI in Geospatial Mapping with Becky Soltanian

Sheikh Shuvo:
Hi, everyone. I'm Sheikh Shuvo, and welcome to Humans of AI, where we learn about the people who are creating the technology that's changing our world.

Today, I'm really excited to share with you Becky Soltanian. She's the VP of R&D at Sanborn, a global leader in geospatial and mapping solutions. She has an incredible career covering lots of different parts of the AI world, and today we'll learn about her and what she's been up to.

Becky, thank you so much for joining us.

Becky Soltanian:
Hi Sheikh, how are you? And thanks for having me here. I appreciate it.

Sheikh Shuvo:
The very first question I have for you, Becky, is if you had to describe your job to a five-year-old, how would you communicate it?

Becky Soltanian:
That's a really good question. And I think about that a lot. Because many people ask me and I have to simplify what exactly I'm doing. If I want to explain it simply, my job is to help, you know– in real life, I'm not working with robots, but I can explain it to someone with a little background in AI that I'm helping robots to see the world as close as possible, you know, the way that humans see. We're not as close, but we are doing our best to, to be that, to get that. And also this is, for example, with the camera, I can show a couple of tricks of how things are working in AI, for example, how a deep neural network works, I mean, show them the graph and everything. It's a little difficult. It's difficult to explain AI in a simple way.

Sheikh Shuvo:
Well, to be more specific, can you share some of the things that you're working on at Sanborn?

Becky Soltanian:
Yes. So, I mean, it's a very historical company, actually. It started, I think, sometime late in 1800 something, and it created, you know, maps for insurance companies. And all these maps were, you know, created manually, everything, you know, some people, you know, they collected data on them, they drew everything, and then gradually beginning of, you know, I think it was 2000 something, the company was acquired by the recent, uh, the current CEO and they drifted toward mostly collecting aerial data and still everything in, in, in the company was, the majority of the things, not everything, was, you know, done by human and manual. So when I joined the company, my goal was to help automate the majority of those works as much as possible and help them with, uh, you know, automating different stuff. And, you know, one of the tools for automation nowadays is AI. And because we are working with imagery, it really helps. And that's the main thing that I'm working on. The company itself collecting data, aerial data, and it's – This imagery has, for example, LIDARs, depending on the customers. And you can, you know, ask for custom even data sets. We can provide that for, for you, for customers that if they request it with some, sometimes they want feature extracted. Sometimes they want some jigs detected. Sometimes they want segmentation and everything. So these are mostly the projects I'm working on, sort of related to this type of projects in AI.

Sheikh Shuvo:
Nice. Now, winding the clock backward, I saw on your profile that you started your career in academia as a researcher. When you made the shift from academic research to industry, what are some of the things that changed for you and how you approach and tackle problems?

Becky Soltanian:
Academia, it's mostly about research. Not always, but most of the time finding new ways and those ways may not have, for example, immediate application for an industry. Sometimes even people work on some stuff that's, you know, sounds like sci-fi. It may, for example, take a couple of years till it hits the industry to use them. For example, AI, I mean, one of the basics of AI is neural networks and neural networks came to the academy. I mean, people started working on that. I think late '90s. 1990 something or maybe a little bit before that and because of the industry wasn't ready and this readiness depended on the hardware I mean there wasn't, for example, those supercomputers that we can use for neural networks or deep neural networks so it remained as research but it took a couple of years I mean decades even till deep neural networks based on that neural network, conventional neural networks started, you know, become a hot topic in industry, in academia, and industry at the same time because of the advancements that were made in hardware industry. So when I came to the industry, it wasn't like that anymore. All the majority of the projects were, you know, demand from the industry. I joined, I mean, as my first job was when I joined, it's a LIDAR company. And the work that I've done for them was analyzing the point cloud and trying to get some of the, you know, some analysis for, for, for example, for the engineers to help them for testing those devices. And, you know, it wasn't like just sitting and reading a bunch of publications and then coming up with something new and writing another publication with code. This was the main difference. It was really, it was a challenge. And at the same time, it was very interesting, you know, to see that what you're working on and there's an immediate demand for that.

People want it. The company invests in that, it's not just, for example, research. And that was the biggest difference that I faced.

Sheikh Shuvo:
Interesting. Well, tell us more about the team that you work with at Sanborn as VP of R&D. What are the different types of roles and people that are on your team?

Becky Soltanian:
We have, I mean, the majority of, I mean, the work that we are doing is coming. I mean, the internal, we have internal customers. Most of the time, for example, other departments like the imagery department or LIDAR department, they reach out, and they want us to work on some stuff. The majority of the people, I mean, everybody is pretty much hands-on coding and, and also, at the same time, have a background in research that helps to, you know, come up with innovative solutions because sometimes there is no solution, there is something close, but not exactly the solution for the problem that we face. We have to go understand that close approach or algorithm, and then come modify it for our purposes. So, that's why I call our team, sort of, not pure R&D, that's, you know, work on publishing. We are applied R&D, that means that we read applications. We think about the application that we have right now in hand, and then we change the algorithm and fine-tune it based on our needs. And there is always, approximately always, there is someone that wants that algorithm as a software tool for them to use it. So that's why I'm here applying to R&D mostly.

Sheikh Shuvo:
Absolutely. Now, with so many internal customers, do you think there are any misconceptions about what you do or something about your work that you wish the rest of your team knew?

Becky Soltanian:
If I understand the question correctly, I think the answer is no because, uh, pretty much everything is clear. And I try to explain and be as clear as possible with my team. So anything that comes, I will share it with them, mostly related to the projects. If something is confidential, then that's another story. But most of the time we are very open about problems, solutions, what we are going next, I mean, what we are going to do next and next steps. That's, I mean, my policy is to be open with the employees.

Sheikh Shuvo:
Good communication usually solves most challenges. It sounds like you're already on top of it. Now you mentioned that your team also tries to stay on top of research in the latest models and things like that at a time in the industry when things are happening so fast, where do you and your team go to learn about the latest and greatest there? Are there particular industry resources, conferences, things like that, that you can recommend?

Becky Soltanian:
The majority of, I mean, I can say that there are a couple of tools that we are using every day, actually, for publications, even sometimes they're, you know, just pre-pub. I recommend arXiv. The majority of the publications you can find, and it's a, you know, free to use platform. There are some conferences, there are famous conferences that we follow them, but not always, you know, helping us because sometimes academia is ahead of us, and it takes some time to get to the point to use their algorithm or their approaches. Then another thing that recently, I mean, we are, I mean, using daily instead of Wikipedia is ChatGPT by OpenAI, and I promise I, I'm not getting any, you know, advertising them. It's just an honest opinion and feedback that we, we use it a lot. Despite that, sometimes it provides, uh, you know, false information, but we use it very often for many things.

Sheikh Shuvo:
What are some ways that you use it that might not be obvious to a more casual user?

Becky Soltanian:
I mean, tools?

Sheikh Shuvo:
Or specifically, just using ChatGPT for your own research purposes. Are there particular types of prompts that you get a lot of value from, particular types of summaries, things like that?

Becky Soltanian:
Well, one of the things is that, for example, my main domain is computer vision. And in computer vision, there are a couple of libraries in Python. The most famous one is OpenCV. And then, you know, for data analysis, people use Pandas. And then we have, for example, this, if I remember, if I pronounce it correctly, Scikit-learn or something like that, and scikit-image from that package also. And then there are some packages in Python dedicated just for, you know, GIS or aerial, like Rasterio. And, you know, sometimes it's using all of these; I'm fluent in OpenCV, but sometimes I'm rusty about using one of the, let's say, functionalities from Rasterio, then I just go to ChatGPT and ask for a piece of code. It's most of the time it works, but sometimes it needs fine-tuning to get the results because, you know, it's, it's an AI model. It's not a human that, or a documentation that provides accurate results. So these are really helpful things. I mean, it helps a lot even to advance the code instead of thinking and scratching my head over something, going to ChatGPT and ask for a small function. And it can do that. And then I can use it and continue. That's become really faster.

Sheikh Shuvo:
Absolutely. Yeah. Shifting gears back to your team, in your current role or earlier in your career, outside of technical prowess, what are some of the softer skills that you look for on your team as you're trying to make a hire there? Are there any particular questions you'd like to ask, any particular exercises you'd like to take candidates through?

Becky Soltanian:
I have a policy, which is a little bit different from other probably managers or something, which is a normal custom in different companies. I believe that someone who studies and gets a degree and enters the industry can learn. I mean, I believe in people being capable of learning. And if I'm mostly, instead of looking at someone memorizing things that in so many companies, for example, they are looking at something like that. In a candidate that memorizes all the functionality of one library can write everything from heart. Mostly I'm looking at whether this person can go search, use, you know, utilize different tools and provide something for me like a basic piece of code or an analysis of a part of data and this is really helpful because people, because I believe that people can learn coding and coding is mostly practicing. Without practice, I don't, for example, expect new grads to know everything about coding or programming, but gradually they will learn how to do the research, how to, you know, come up with an innovative algorithm or change something in order that they can use it. Fine-tuning an algorithm. This is something if I really appreciate in, in, in a candidate and I really look at something like that instead of how fast the person can write a code.

Sheikh Shuvo:
Absolutely. Now that's great advice there. The very last question I have for you, Becky, is, um, putting your venture capitalist hat on. There seems like there are a thousand new AI companies every single day of building all sorts of different tools from your. Knowledge of the different tools and problems out there. Is there some of, uh, AI problem that you have that you haven't seen a solution for that you'd love to buy a tool to accomplish that?

Becky Soltanian:
Of course. I mean, I have a wishlist that I want a computer just, you know, I mean, this is a joke, by the way, I want a monitor that when I stare at it, it solves the problem, which is in my head, but this is not happening yet. So it's. Yes. Sci-fi may not happen. it would be nice in the future we see something like that. Yeah. I, I think nowadays AI just, you know, takes over the majority of the areas, but some of the areas that I see that there's not, I mean, they ha it, it hasn't been touched that much is, for example, real estate. Mm-Hmm. Or architecture. Architects. Majority of the things are still, I, I believe that those tools are not automated. Human involvement is a lot, many other, for example, in those days, like small manufacturers, they are not, you know, using AI that much, or even not at all, because, you know, they're small businesses, but gradually they have to start thinking about that and accommodating AI in their daily routines. These are, I think the areas that some real areas that haven't, I mean, AI didn't reach yet. Uh, it will in the future. So I think that's, that's something that I can advise people that they can explore and work on.

Sheikh Shuvo:
Sounds like fun opportunities to tackle there. Well, Becky, that's all I have. Thank you again so much for your time. And anyone has any questions for you? What would be the best way to find you online?

Becky Soltanian:
I'm on LinkedIn. It's Becky Soltanian. They can search with this and then they can send me a message, write me comments and on under my post and they can follow me.

Sheikh Shuvo:
Awesome. Thank you again for making time.