Adaviv: Bringing Plant-Level Data and Insights to Agriculture

[MUSIC PLAYING]

IAN SEIFERLING: Hi. My name's Ian Seiferling. I'm CEO and co-founder of AdaViv. We're a precision agriculture platform for controlled environment farms, looking to use our machine vision AI technology to help them essentially be more profitable by lowering input costs and automating much of the tedious tasks that happen in the cultivation.

Sure. Well, I suppose the company started back with my roots. I grew up in Saskatchewan in Canada, a farming province. So every family had a farm in it at some point. And so I just grew a really strong connection to agriculture, to nature, followed that through my PhD in environmental sciences and biology, and ended up at MIT during my PhD. I was lucky to have that opportunity and stayed on for a couple of years as well as a postdoc.

There I was researching urban agriculture, looking at how scalable it is to grow food in and around cities. And one of the great things about working in an applied university and a lab like the Sensible City Lab where I was was I got to work with many brilliant people from across disciplines from biology to math to AI to physics to design.

I met my co-founder Moe there. We work together as postdoc researchers. And it was then that we got this passion for entrepreneurship that MIT helped support. From my work in plant science, I knew that these powerful new technologies, like AI, machine vision, data science, were really ripe and ready to help how we optimize how we grow crops in controlled environments, essentially by giving us the ability to collect plant level data, which allows us to close the loop between the plants, and the control systems, and the farming systems, and the inputs, and really understand those feedback, and run a much more lean, optimized, essentially production system, which a farm really is.

So we worked together on that, leveraging our experiences-- Moe's as an AI scientist. And we quickly met Julián, my other co-founder, who was completing his MBA at Sloan MIT. And he really brought a lot of experience working with McKinsey in lean manufacturing. And we really felt that was a really important piece to what we were building in this platform to be able to connect the data to actual value creation in terms of lean efficiencies, lean workflows, and really put that plant level data into action to help farms both get higher yields, but do it with less costs and less inputs.

So again, we work with controlled environment farms, things like greenhouses, indoor farms. They have the ability to bring more control to optimize water inputs, energy. But these farms are still traditionally run with a lot of subjective decision making, and are still relying on people's ability to get into the field and look at the plants, look at the condition, detect pests disease pathogens by human eye. And that's really tough because the scale of these farms is massive. Labor is expensive.

So basically, they lack a lot of visibility over what's happening in the farm, especially the managers or the cultivators that have all of that expert growing knowledge. They don't have the time. They don't have the bandwidth to be in the field to be giving personal care, so to speak, to every plant.

And still, a lot of this is done, communication, decision making, on pen and paper. And so what we're doing is quantifying better that plant level data. And we're detecting issues much, much earlier and at larger scales than they can do with traditional methods. We do that by using machine vision to collect that plant level data to detect pest disease pathogens, but also measure the growth of the plant and indicators that the growers can use to make better crop planning decisions and to evaluate what their team is doing in the field.

So what we do is, A, collect plant level data, use AI to turn those images into key metrics, and detect issues much, much earlier. We detect 7 to 10 times more issues than they typically would capture with traditional human methods, automate that scouting, and we help them quantify plant growth. That helps them inform yield predictions and define how they're going to optimize, or plant the crop, or-- excuse me-- help them optimize their growing recipes, essentially, because we are collecting that data that tells us how the plant signals, telling us what the plant is wanting, and needing, and reacting to the environment.

So we help them bring that together in one single platform. And that's why we call it a lean cultivation platform because it's not just about machine vision. It's not just about data. It's about turning that data into workflows that help the team execute better, reduce waste. So at the end of the day, we're helping them increase the yields, but also doing it with less input costs.

I think what I mentioned earlier, that sometimes we are able to identify these patterns and trends that get lost in the complexities of the day to day, or get lost in the complexities of running multiple crops at the same time. And that is trends where we see certain pest or disease issues start and spread. We can often trace those spatial patterns back to where they came from. Did they come from a certain area of the farm?

And so those insights are super valuable because they're more or less invisible to the managers, because they don't have that oversight over the farm. So identifying spatial patterns is one. Identifying temporal patterns-- is there seasonality to the issues they're facing? How can they use that information to plan the next crop better. If we know that we're in a season where pathogens are extremely high, where we have maybe a section of the farm that is prone to high humidity and pathogens that we've identified, they can then grow only cultivars that are maybe more resistant to that pathogen in that section, and so that they're optimizing what they grow to get the most out of those plants. So that's one-- identifying those kind of invisible spatial and temporal trends.

The other is sometimes I think often growers find it hard to quantify how much they're spending on inputs and quantify how much certain tasks are taking. So some farms or some crops, they do a culling process early in the cycle of the plant, whether that's a clone, or whether that's a germinate. They want to identify plants that are growing, that are performing well, and plants that are not performing well so that they can remove those from the crop when they move to the vegetative and the fruit producing stage. They're optimizing those best performing plants. So it's an optimization.

And it's extremely labor intensive. And they often don't know are they spending too much time, and which are they really using data-driven or quantitative metrics to understand what is a poor performing plant and what is a high performing. We give them quantitative data on that. So it goes from subjective to objective. And we can automatically detect the, say, 10% plants that are not growing at the same pace, and make that process go from an hour-long process to minutes, and do it much more accurately.

We're constantly innovating, constantly developing additional features. This technology is very flexible in that manner. I would say three things we're very excited and focused on right now. One is thermal imaging. Two is additional automation. And three is partnerships.

So with thermal imaging, traditionally, that type of sensing is very expensive and hard to do at scale in the field. Maybe in a lab setting it works. But we have now deployed very, very high resolution thermal imaging into our crop scanning system. And that gives us the ability to see plant stress, to see plant performance indicators and metrics that are completely invisible to the human eye. So we can understand when a plant is stressed days before a human could ever detect if it's got an issue, like a fungus, or a bug, like aphids.

And that's a game changer, really, because the earlier you can detect crop issues, of course, the better. And you're able to prevent the spread. Often, if you see it too late, it's already spreading. The populations are growing, and it's very hard to get under control. So we've already had quick wins there.

It also helps us understand when is the plant working. We can tell when it's transpiring, when it's not, and the growers can use that information to optimize their environment and their growing recipes so that the plant is working when they want it to work. And it's kind of resting when they want it to rest.

The other, as I mentioned, partnerships and automation. Partnerships are key in agriculture Today is still very fragmented in terms of the technology space. And we're unique, I think, because we generate this plant level data, this fundamental unit of the production system.

And a lot of other players in the space, the traditional leaders in the market, in the lighting, in irrigation, in nutrients, in greenhouse supplies, in design, they're all asking and wanting more data to drive more actuation and automation in their systems. So how can we fine tune our light recipes? Well, we need to know how the plant responds and be able to create those quick feedback cycles. And so we're talking to a lot of these leaders now in the industry because we can provide that information and that intelligence layer that they're all seeking.

And then, lastly, is automation. What we've built today, we've shown a lot of wins-- increased yield, reduced labor, more efficiency in the cultivation team. But we want to increase the automation. And we want to work with more players in that space, too. Agricultural robotics is a very quickly growing sector. And there's multiple tasks that AI or robotics technology are performing right now, and we want to see those come together. So we want to leverage our plant intelligence platform to integrate with, perhaps, a harvesting robot, for example, to provide a more full solution, full automation in terms of automating those kinds of manual tasks.