LAWRENCEVILLE, Ga. — At 14 years old, Sirish Subash is now America's "Top Young Scientist." The 9th-grade student from the Gwinnett School of Mathematics, Science and Technology won the national competition for his invention of PestiSCAND.
Built to use with a phone application, the PestiSCAND device uses machine learning to detect pesticide residues on produce.
11Alive had the chance to interview the "Top Young Scientist" and peer into his mind and thought process when creating PestiSCAND.
Scroll below for the Q&A with Sirish Subash.
What was the inspiration behind this? Why pesticides?
This, this problem came up when I was talking with my mom over some produce that we were washing. And my real question was, 'Why does it matter that we wash them? And how effective is the washing process?' So I researched, I did, I looked into some papers and I found out that you don't pesticide residues can stick around after multiple rounds of washing. And they've been linked to all sorts of health issues. They're found on most of the produce that we actually eat.
So, if we could detect them, we could make sure that we weren't consuming them. And that's really where this whole project started.
When did you come up with this idea? How old were you?
So I came up with this actual idea about a year ago, maybe roughly a month back. So towards the beginning of October and around that time frame, I was just working on figuring out how might I actually approach this? And I settled on the method that I'm using right now.
So you have a spectrometer. What does that do?
That's this sensor right here. And what that does is it basically shines different wavelengths of light that's ultraviolet visible and infrared onto the produce and different chemicals based on what they're actually made out of. Different chemicals they have different spectral signatures, which is how they reflect different wavelengths of light, what wavelengths or parts of light they reflect back. So PestiSCAND can look for the spectral signatures of pests that residues on produce.
And this is something where it's an app you built on your phone. Tell me how that works.
So, the actual sensor and the Bluetooth transmitter on this piece here, and they connect over Bluetooth to a mobile app. And the app is something that this is, this took most of my time this summer developing and building the app and what the app does, it'll take the data, the spectral signatures from the sensor. It'll feel it, it'll feed it into machine learning models and basically what those do is they'll take the data and they'll turn it into the results that you'd see on the screen.
Are there pests that residue present? And which pests that residues are present? So you had to code that in; I'm thinking of, a whole coding sequence, right?
Yeah. So, there's a few key parts here. So there's the actual getting the data and transmitting it over to the app. And then there's one of the arguably more important parts which is taking the data from the sensor and then getting the results out of that. So, these were the machine learning models doing that. And what I did was I took a bunch of samples of produce with pest set residues and produce without pesticide residues. And what I did was I trained a machine a computer to look at that and find patterns in the data.
Looking at what can you commonly find for produce with pesticides residues versus produce without pesticide residues. And using that, you can take the data from the sensor and turn that into the results search.
What pesticides does it detect?
It's been trained with three separate pesticide residues. Now, the reason I chose those is because they were some of the most commonly found produce item pesticide residues on produce.
What produce does it work on?
So far, it's been trained for apples, spinach, strawberry and tomatoes. So it's, yeah, spinach here is one thing was I had plenty of it as well as it's one that can have pesticide residues on it. And it's also a part I've read on some, some of the research that I've found that spinach is also one of the most commonly found.
Did you also do the 3D printing of the machine itself? Can you talk a little bit about that and just the programming of it?
So, the actual physical structure of the device is mostly 3D printed. There's a couple other parts in here, but the main, most of the structure is 3D printed. I designed these all myself and printed them out. And for the actual programming, this was one of the bigger parts of the actual project. But over the first few rounds, I started out with a couple of programming which just being Python and CC or C++, it's a site variant on both.
And right now, I've shifted over to Swift, which is a language for developing IOS apps and I'm still keeping in the C or C++ component of it for the controller board here.
And what is the guts of it? Is it like an Arduino or Raspberry, or what's working behind the scenes?
This is the actual controller board here. It's similar to an Arduino, it's called a red board. And the reason, the main reason I chose that because it had easier compatibility with the sensor and it had Bluetooth integrated so I could connect to the mobile app.
That's a lot of programming languages to know at your age. Did you learn some of those in school?
So while working on this, the most of the languages were I really started out programming with Arduino and there's a C C++ setup as well as some Python that I learned earlier on. That's really what got me into programming. And then I also worked with the Swift here. This is a language I learned for the project. I really just did a lot of research into the language and what components I would need to know for this project.
At what point did you feed all the information into artificial intelligence, and then you see it taking over learning what you wanted to? How exciting is that in your head that you've kind of created a monster almost?
I mean, I think it's interesting and it's really saved me a lot of time because it, it also, it's also made the device a lot more accurate. So if I had simply hard coded in bound, like say, if this value is greater than another like threshold, then it would be far less accurate and wouldn't be able to pick up on these smaller minor changes that are happening in the data between produce with and without pesticide residues.
With the machine learning and AI component this did is by feeding it this amount by by feeding it this data, I can get the actual device to pick up on small, more small, more nuanced details to help it determine whether pest set residues more accurately.
So, you think machine learning is a helpful thing for technology developers?
Yeah, it's really helpful because it allows us to pick up on details that normally just the human eye wouldn't be able to. And it also speeds up scaling things. What this means now is instead of manually looking at all the data for different proto items, I can give the actual machine learning models, different samples. So if I simply scan different produce items with different pesticide residues on them and upload that data to the models, they can pick it up from there. As opposed to me actually looking through the data for each proto set.
There's been food that has been found to have E. coli in it, and they've had to recall some food, any thoughts of taking it in that direction or is that even applicable to this hardware?
One thing that really makes this sort of technique or spectrometer really usable here, more versatile is that it can be adapted to finding a vast variety of different materials and chemicals. So that could be a part of a potential application for the future.
Is there anybody in your classes who's tried to develop anything on the same level? Like in a different way, you developed this application? Are other people developing other machine learning in your classes?
So not in particular that I that I know about, but a lot of my other classmates are still, they're working on different projects like this but not in particular similar to PestiSCAND.
You said you did 3D printing, but where did you get the printer? Do you have one at home, or was there one here?
Yeah, I had one at home that I still got is we got a couple of years back and this is the first time I've majorly used it for something.
How long did it take you to your first?
The first, I think it took, it took three or four months from September to October that time to late January to late December.
And then you just kept modifying like small tweaks every time?
Yeah. So, the first working prototype was by December. I made some other versions before that, but those didn't really work out. And then getting to this prototype, I really started to get to work in like late May. As soon as school was going to go out over the summer, I really worked on that project with the first half of my summer really being spent for the app and getting the app developed.
And then after that, it was starting in mid-July working on the machine learning models and then getting the actual physical device prepared.
I'm so impressed like you did that as an eighth grader, were there classes you took in middle school to help you learn how to do that or did you just study on your own?
So there weren't any particular classes in middle school that I took. But I really spent some time like after school on weekends, particularly researching, seeing how I could work on this and doing stuff after school mainly.
When you've been looking to study outside of school, were there any more topics that you studied, or podcasts you listened to or programs you saw online that you were like, "Wow, that's intriguing?"
Yeah. So, I listened to this one news podcast called Kid Nuz. And one thing that, that showed me was also to these the current events. But also I heard about the 2017, Young Scientist Challenge winner Gitanjali Rao. When I heard about her project, that was really what sort of set me off and what showed me about the Top Young Scientist challenge.
And that's really what told me that this was kind of the way to go for me because I saw her project is something that would help solve and help improve lives. That's really what I wanna do with projects like this.
So when you knew it worked, you said, "OK, I want to enter this competition." What was the path of preparing yourself for going to the big competition earlier this month?
Yeah. So, for this month's competition, that was the final event with the 3M young scientist challenge for that. The preparation was, it wasn't too much. I mainly spent some time every day looking over my, what I'm actually gonna be presenting and speaking about just presenting my project practicing that we have a five-minute presentation, we have to give with a five-minute Q&A. So, just prepping for that over and over and over again to get it down.
How did you practice that? Because one of the 3M contests represents brilliant discoveries and research but presentations that are very tangible. How did you practice that at home?
So, making that makes sense to people. I really started out with just breaking down what I was trying to say and looking for any jargon words, anything that could get in the way of it being understood. Another thing I did was I practiced with my sister who is in second grade now and she really helped me. She just pointed out little things that I could change just to make it clear and overall a better presentation.
So, your biggest critic?
Yeah, she was my sister.
I love that. So you won a huge prize, $25,000. What does that mean to you?
I'm really excited about it. And what I really wanna do with this is I wanna put it towards college and with that, I just wanna push innovations like PestiSCAND further so I can help solve real-world problems.
And do you already have companies reaching out to you to try to buy it?
Not quite yet. But I'm working on that right now. It's a prototype, but I'm working on getting out to a form where eventually it could be put on to the market.
Did your classmates react when you won?
Yeah, they actually read about it like on social media. So they were all like talking about it. They were all excited as well.
And for younger kids than you, maybe second grade, third grade, fourth grade. And they see that you invented this and you went to the competition and you won. What do you hope that they can see?
Well, the biggest thing that I hope they can learn is that what, what really hasn't been shown but should be is that there's a lot of failures along the way. So this. It wasn't really just a one and done. There were a lot of different stages in between where I started and where I got now and just by having grit and perseverance going through and just keeping going after all the times that I've messed up with this project I through that was how I was able to get here. And I just like to convey that message of stick to it and have perseverance.
And now you're just in ninth grade, you got many years ahead of you before you graduate. What do you have any idea what you wanna do when you, when you grow up and go to college? Like when do you wanna study further in life?
I wanna study particularly physics. Physics is a field of science that really interests me. But I wanna look at with the mindset of applying it to life and making the world just designing innovations to solve real world problems.
So physics, physics gravity, or do you mean circuits?
I feel like particle physics really satisfies my curiosity, like on the smaller side of things quantum physics is something that I enjoy, but I'm still yet to decide on like a specific sub field of physics.
You should create a YouTube channel where you teach people about really complicated things because you're so good at breaking it down to tangible!
I've been, I've been having a YouTube Channel for the past few years. It's been a minute since I last posted a video since I've been busy with projects like pasta scan.
So, what kind of videos have you done on that before?
Mainly like science explainer-type videos. But yeah, that's, that's been, you already did it.
And that's why you're so great at the 3M! What's your YouTube channel called?
It's called SciKid Sirish.
Yeah, seeing the successes from this. Where does your mind go next?
So, I really, I haven't decided on the next challenge tackle just yet, but I'll be, I'm just on the lookout working on things like this, deciding on just solving any other problems that stick out in the real world.