Computer science master’s student is helping develop AgriScout, an autonomous crop-scouting platform designed for real-world agriculture
When Jacob Serafin began studying computer science at the University of Guelph, he imagined a future spent largely behind a computer, writing code and solving software problems.
He did not necessarily expect that computer science would have him building robots, working with electronics, learning to weld and testing technology in farmers’ fields.
Now a master’s student in computer science, Jacob is part of the team developing AgriScout, an autonomous crop-scouting platform that uses artificial intelligence and computer vision to help identify crop disease, pests and other signs of plant stress.
The project has pushed his work well beyond programming. It has also given him the opportunity to see what happens when research leaves the lab and has to perform in the unpredictable conditions of the real world.
“I think the biggest change is how I’m much more comfortable taking on problems where I don’t already know the answer or even have all the skills I need to start,” he says.
Using AI to give farmers a closer look at their fields
Jacob’s path to agricultural robotics began during his undergraduate degree at U of G, when he met Dr. Gurjit Randhawa through one of his final courses.
After the course ended, conversations about Randhawa’s research introduced Jacob to challenges in agriculture where his computer science background could be applied. What initially interested him as an AI project quickly grew into something much broader.
AgriScout travels through agricultural fields capturing images of crops. Those images can then be analyzed using AI models trained to recognize signs of disease, pests and crop stress.
The goal is to turn that information into a detailed picture of what is happening across a field, helping farmers identify potential problems and make more informed decisions
“It’s an autonomous crop-scouting platform,” Jacob explains. “It travels through the field imaging crops and then we use AI to identify signs of disease, crop stress and other conditions that may be difficult or time-consuming to monitor manually.”
One area the team is investigating is disease detection. Identifying where disease is developing could eventually help farmers respond more precisely rather than treating an entire field in the same way.
For Jacob, that is where the value of the technology becomes tangible.
If we’re able to detect disease early, we can stop treating the field the same and allow farmers to treat disease hot spots instead of blanket spraying the whole field, he says. It allows us to not only reduce crop loss but use farm inputs more effectively.
Taking computer science beyond the computer
Jacob initially expected his contribution to AgriScout would centre on computer vision: collecting and processing images, building datasets, training disease-detection models and getting those models running on the robot.
It did not stay that way for long.
As the project developed, he became involved with cameras and sensors, computing hardware, CAD, electronics, manufacturing and mechanical components. He even learned to weld.
That experience changed his understanding of what it takes to build an AI system.
Before I thought the AI model was everything, but it’s actually not, he says. It’s everything else that goes around it.
Developing an accurate model is only possible if the team can reliably collect good data to train it. Cameras have to work under changing light. Electronics need sufficient power. Mechanical components have to withstand uneven terrain. The robot needs to operate when connectivity is limited. And agricultural fields are not controlled environments.
Changing sunlight, dust, rain, wind and rough terrain can expose problems that would never appear during testing at a desk.
Sometimes those lessons have arrived dramatically, including broken robot legs on uneven ground or a battery dying in a field at night.
“You have to think about cameras, power, electronics, mechanical design, networking and how you can collect good data,” Jacob says. “Moving research into the real world means we need to stop asking whether what we develop works and start asking whether it works for the person who is going to use it.”
Learning across disciplines at U of G
AgriScout has also introduced Jacob to a collaborative research environment that extends far beyond computer science.
The multidisciplinary environment has made me a much more open-minded researcher, he says. It has taught me that some of the best solutions come from people who look at the problem completely differently than I do.
Those relationships have made learning a shared process. As Jacob develops new skills from other members of the team, he can bring that knowledge back to the lab and pass it on to students joining the project.
“You don’t need to be an expert in everything, but you do need to know how to work with and learn from people who are.”
His experience at U of G has given him the foundation to do that. Coursework helped him build knowledge in areas including software engineering, AI and data analysis, while research has challenged him to apply those skills to problems without predetermined answers.
My education at Guelph gave me that foundation, he says. The combination of coursework, mentorship, research opportunities and actually getting my hands dirty by building things, breaking them and fixing them has been one of the most valuable parts of my experience at Guelph.
Building, breaking and finding another solution
Working on AgriScout has also encouraged Jacob to look for solutions in unexpected places.
“Sometimes I’ll walk through Home Depot aisles and just look at parts and think, ‘How can I use this?’ or ‘How does this part work?’” he says.
One solution came from even closer to home.
At one point, the team needed a material with the right properties for a component inside the robot. The solution turned out to be a cutting board.
There is literally a cutting board from the kitchen inside the robot, Jacob says. We needed something with the right material properties, and the cutting board just worked.
That willingness to experiment is something he traces partly to his family. His father’s friends use the Polish nickname złota rączka, literally “golden hands,” for someone who can build or fix almost anything.
Jacob likes to think some of that has made its way into his own approach to research.
“For me, there’s something incredibly rewarding about starting with an idea, building it, working through all the problems and then seeing it actually working in the real world.”
Research designed for the real world
Working directly with farmers, researchers and industry partners has added another dimension to Jacob’s education.
Their questions and feedback can quickly reveal whether an idea that makes sense technically will actually be useful in practice.
For technology like AgriScout, their perspective is essential. A system cannot simply be technically impressive. It also has to make sense for the people expected to use it.
“If the system is too expensive, difficult to operate, unreliable or just simply doesn’t provide enough information that helps them in running their operation, it’s probably not going to be adopted.”
Seeing the project used and evaluated beyond the lab has made his research feel more tangible and reinforced what drew him to the work in the first place: the opportunity to use computer science to build something with practical impact.
It has also changed what Jacob sees himself doing after graduate school.
Once you start building something like AgriScout and seeing the problems that you can solve with AI and robotics, you start noticing opportunities everywhere, he says.
Whatever the next project is, he knows he wants to keep building.
“I really like the process of taking an idea, building something useful, testing it in the real world and hopefully turning it into something that has a meaningful impact,” he says. “AgriScout definitely won’t be the last thing I build.”