“There’s a huge problem with agtech”, says Sumer Johal.
“Everybody’s trying to build applications with a huge amount of their stack being infrastructure; everybody has this huge burden of custom code they have to write.”
“It reminds me,” muses the India-born executive director of the AgStack Foundation, “of the days when people used to build applications before Operating Systems: you’d get an application to directly connect to the disk drive and have to work with the printer and an i/o device like a keyboard.”
“Then Operating Systems came through,” he spells out, “and apps were able to be built with much more flexibility; they didn’t have to worry about that interoperability. That created an explosion of apps and an ease of not just creating them, but an ease of consuming them…”
Johal, who moved to the US from his native India for college “grew up in an agricultural family”, but opted to study computer science at both undergraduate and post-graduate level. His heart now lies with bringing the best of both worlds together: making the abstractions of machine learning, for example, translate into real-world improved outcomes for both farmers and consumers.
The founder and CEO of Agralogics, a company dedicated to using Big Data to optimise farming outcomes, is now trying to tackle this “huge problem” of highly atomised agtech innovation through the Linux Foundation, which this month (May 2021) became home for the AgStack Foundation — after time and again running into the issue of completely incompatible agtech applications.
Enter the AgStack Foundation
The AgStack Foundation launched in early May with the aim of helping improve global agriculture efficiency via open source and open data, and has brought in a large community of interested organisations. (“AgStack’s goal of creating a shared community infrastructure for agricultural datasets, models, frameworks, and tools fills a much-needed gap in the current agtech software landscape,” notes Michael Stenta, founder of farmOS.)
Lest anyone wonder just how much technology is getting used across fields, paddies, orchards, and beyond, the answer is a lot. Juniper Research suggests agtech will be a $22.5 billion market by 2025 – spanning everything form agricultural sensors for crop management, GPS field mapping, and supply chain management, with the ultimate goal of increasing yields and reducing costs through connectivity and data insight; the global agriculture ecosystem itself meanwhile is variously valued at between $5 trillion and $25 trillion.
Johal says: “When you talk to many agricultural stakeholders, they’re pretty savvy people, they may not be your stereotypical tech people, but they’re pretty savvy. And one of their biggest challenges is that they’ve got maybe 20 apps that they use, and there’s no interoperability between them.
“This doesn’t allow competitors to actually work for a common stakeholder like a farmer. And digital technologies promise tremendous improvements to both efficiency, but also better decision making driven by data.”
There’s no shortage of techno-utopianism out there and with farmers grappling with supply chain, procurement, weather, or disease-related issues, are apps and more fiddly, time-consuming tech really the answer?
We put the question to Johal. He’s quick to respond: “Information is a great equaliser. [With better data] farmers can make better decisions as to what to plant when to harvest, how to irrigate, you know, how much nutrient to apply, when to apply it, all those things have massive impact… And one of the key tenets of what we’re trying to create is free infrastructure that anyone can use, whether they’re a member of AgStack or not.”
Among those involved is Hewlett Packard Enterprise, whose Janice Zdankus, VP, Innovation for Social Impact, Office of the CTO noted in a AgStack release: “The world’s food supply needs digital innovation that currently faces challenges of adoption due to the lack of a common, secure, community-maintained digital infrastructure. AgStack – A Linux Foundation’s Project, is creating this much needed open source digital infrastructure for accelerating innovation. We at Hewlett Packard Enterprise are excited about contributing actionable insights and learnings to solve data challenges that this initiative can provide and we’re committed to its success!”
So what is this planned infrastructure, exactly? “We have four large pieces that interconnect”, Johal tells The Stack. “They can be used in isolation, but they also interoperate. The first is data: there’s a tremendous amount of data, weather data, satellite data, soil data that’s already public and free and open. But to go hunting for it requires a lot of knowledge, special expertise, and you’re constantly trying to translate what the heck it means.
“[We aim to] create a simple credential and access point for these data sets, so that with one user ID login, or an API, users can log in and get access to it and translate that to their units, their language, etc.: having have a hub where all these datasets can essentially dock.
“The second part is models: there’s a lot of models or numeric formulations out there covering how pests evolve, or how crop nutrition happens, or how soils work or even how a tractor shaft works. That code is actually really important, because otherwise you’re out hunting for a paper that some Professor wrote; trying to understand the math and then trying to create code for that.
“Why not have a common repository, a code base that is searchable, that is dockable that people can contribute to and have it all in permissive licencing. Having that code base can allow both researchers and also code application builders to very quickly deploy models in their application without having to develop the model themselves. This repo would work with the data structures, so that these models can be all automatically instantiated with weather data, or soil data or satellite data, etc.
“The third piece is what we’re calling frameworks: these are things like user registry, asset registry frameworks that will… give you, say, polygon boundaries of field boundaries in Nairobi. We could basically create and maintain field boundaries at a global scale, which is a massive ask. But again, this is what is needed! Then an application developer can start using that field boundary’s native address to get data for that address from the data repository. These inherent data structures will need a host, so will be hosting that.
“The last thing is extensions and toolboxes: there’s a whole bunch of things like CAD image libraries that are used by machine learning algorithms to compute whether something is a piece of lettuce or an almond, etc. Those ML models need tagged image datasets, or tagged audio datasets — which are really useful to develop applications, because learning has to happen off of something that already exists. The idea in short is we’ll take these four things, sew them all together, so people can do both creation and consumption faster, and cheaper.”
Sumer Johal’s mind is racing. He’s not alone. AgStack has brought in a formidable membership already. The NIAB crop science group is among them. As Dr Richard Harrison, Director of NIAB Cambridge Crop Research puts it: “Climate change is a global problem and agriculture needs to do its part to reduce greenhouse gas emissions during all stages of primary production. This requires digital innovation and a common, global, community-maintained digital infrastructure to create the efficient, resilient, biodiverse and low-emissions food production systems that the world needs. These systems must draw on the best that precision agriculture has to offer and aligned innovations in crop science, linked together through open data solutions. AgStack is creating this much needed open-source digital infrastructure for accelerating innovation.”