At Zoho’s first From the Ground Up AI Hackathon, five student finalist teams met in Pleasanton, California, on September 22 with a challenge: use AI to address environmental problems affecting rural communities.

What they built was strikingly specific. Instead of trying to “solve water scarcity” in the abstract, the finalists built systems around decisions people already have to make: which drought scenarios deserve deeper modeling, where water may be leaking, when and how much to irrigate, how to preserve knowledge about a farm across generations, and where scarce irrigation water should go first.

That specificity is what made the competition matter. From the Ground Up showed that community-centered AI can produce focused, practical systems when development starts with the people closest to a problem and the decision they need help making—not with the technology itself.

I joined ecommerce entrepreneur and ColderICE CEO John Lawson III and AI ecosystem builder Ash Kumra on the three-person judging panel evaluating the five finalists.

It also made the judging much harder. All five teams had identified real problems and built credible responses, which meant the judges had to find meaningful differences among projects that were all worth taking seriously. As we pressed on feasibility, data, deployment, adoption, sustainability and technical resources, the question shifted from “Is this a good idea?” to “Which project is most ready to keep moving?”

Dos Ojos ultimately won the $15,000 prize. Twin brothers Edgar and Eduardo Bello Gonzalez combined lived experience with drought, a technically credible satellite-and-drone approach, a simple bilingual SMS delivery system and newly emerging university support. That combination gave judges greater confidence that the project had a credible path to continue developing.

But the larger lesson was visible across all five finalists. For farmers, utility workers and planners, the most useful AI was not necessarily the system that looked the most sophisticated. It was the one that turned technical complexity into a clearer next decision.

From the beginning, From the Ground Up was structured around a constraint that is easy to overlook in AI development: a system can be technically impressive and still be a poor fit for the people expected to use it.

Zoho invited students from designated rural and rural-serving colleges in California and Texas to build around environmental problems affecting communities like their own. The premise was that students who lived closer to challenges such as water scarcity, wildfire risk and agricultural instability might understand details, tradeoffs and day-to-day decisions that someone approaching the problem from farther away could miss.

That proximity mattered because the competition was not simply asking students to apply AI to a broad category like “water scarcity.” It was asking them to think about what technology would actually work in communities where connectivity, money, technical staff or computing infrastructure may be limited.

Zoho organized the competition around five principles: community sovereignty, appropriate technological scale, interoperability, equity of access and honest environmental accounting. Bigger was not automatically better. A smaller model, locally operated system or simple interface could be more appropriate than a technically elaborate platform if it solved the problem without creating new barriers for the people using it.

The judging criteria reflected the same philosophy. Technical execution counted, but judges were also asked to evaluate problem definition and community impact, real-world validation, feasibility, scalability, sustainability, innovation and how well teams communicated and adapted during questioning.

Those categories became increasingly important during the final because the prototypes themselves were strong. Once a team had demonstrated a compelling idea, the next questions were harder: Where would the necessary data come from? What would deployment actually require? Could people use the system with the resources available to them? Who would continue building it? And what would allow the project to keep going after the hackathon ended?

Those questions did not diminish what the students had built. They revealed what each project would need next.

And across the five finalist teams, the answers were very different.

Texas A&M University-Corpus Christi students Noah Wilborn, Mihail “Misha” Stegall and Mohammed Asad Khan formed team NoNiMo and approached water scarcity from a planning problem that can become expensive quickly: deciding which possible drought scenarios are worth modeling in depth.

Their system, BASIN, used public precipitation data to generate and rank drought “what-if” scenarios. The goal was not to replace formal hydrological modeling. Instead, BASIN was designed to help planners narrow the field first, identifying which scenarios appeared important enough to justify the time and expense of deeper analysis.

That distinction made the project practical. Rather than asking AI to make the final decision, NoNiMo used it to help determine where more sophisticated human and technical resources should be spent.

What made the project even more striking was how little time the team had to build it.

While the other finalists had weeks to develop their systems, NoNiMo told judges that university requirements left them with only about three days to put BASIN together. Before contacting outside experts, obtaining certain data and taking other steps that would support the project, the students said they had to work through institutional approvals similar to the kinds of constraints they could encounter in real-world research.

Despite that constraint, the team came into judging with an unusually complementary mix of strengths. The team brought a combination of computer science and cybersecurity expertise, with Wilborn’s cybersecurity background adding another layer to how they thought about deployment and risk. Khan told judges that his existing work gave the team access to satellite and other datasets they could need as the project developed.

That combination helped answer questions other teams were still working through: Where will the data come from? Who has the technical expertise to continue building this? And what happens when an AI system moves beyond a prototype and begins interacting with consequential infrastructure? Wilborn’s cybersecurity experience meant NoNiMo was already thinking about how a system connected to real-world water infrastructure would need to be protected from outside threats.

BASIN came very close to winning. Its combination of technical capability, access to data and early cybersecurity thinking gave the judges confidence that the team understood not only what it wanted the AI to do, but some of what would be required to make the system usable beyond the competition.

Just as importantly, BASIN kept the final judgment with people who understood hydrology. The AI could reduce a large number of possible scenarios, surface the ones that appeared most important and help hydrologists ask better questions. But experienced human experts would still determine which scenarios deserved deeper modeling and how those results should ultimately be interpreted.

Given the roughly three-day build window the team described, that was an unusually mature division of labor between AI and human expertise.

Kuralay Biehler and Lazzat Duiseshova of Austin Community College-Round Rock approached agricultural instability from a problem that technology cannot easily recreate once it disappears: the accumulated knowledge a farmer carries about a piece of land.

A farmer may know where flooding has occurred, which areas suffer first during drought, what interventions worked, what failed and how different parts of the property have changed over decades. Much of that knowledge may never exist in a formal database. When a farmer retires, dies or passes the land to another generation, some of it can disappear with them.

Land Memory AI was designed to help preserve that history. The prototype used a familiar, Instagram-like scrolling interface where farmers could add photographs, written observations, voice notes and information about previous events on the property. The idea was not to replace farmers’ judgment. It was to make the knowledge behind that judgment easier to preserve and pass forward, so farmers could look across years of experience, identify recurring patterns in good and bad seasons, and use AI to surface trends that might otherwise be difficult to see. That historical context could then help farmers make better-informed decisions about what to expect and how to respond.

The simplicity of the interface was deliberate. Biehler and Duiseshova told judges they were designing with older farmers in mind, including people in their 60s who might not want to learn an entirely new technology. Buttons were large and easy to identify, photos could be uploaded directly from a phone, and the scrolling feed resembled social apps users might already recognize. Farmers would not need to buy specialized equipment or learn a complicated new system simply to begin recording what they knew.

For Biehler and Duiseshova, the problem was also deeply personal. During their presentation, both women told judges about family histories involving generational farmland in Kazakhstan that their families had lost. They had not known each other there, but after meeting in Texas, they discovered that they shared remarkably similar family experiences.

Their presentation was one of the most emotional moments of the competition. It also helped explain why their understanding of the problem felt so specific: they were not simply building a database for farmers. They were thinking about what can be lost when knowledge tied to land, family and generations has nowhere to go.

The team also imagined Land Memory AI eventually drawing on information from neighboring farms. With enough shared history and current environmental information, they believed the system could potentially help communities recognize patterns related to drought, flooding and other events. That remained a future direction rather than something the prototype had already demonstrated.

Judges repeatedly praised how well Biehler and Duiseshova understood the community and problem they wanted to serve. But their project also exposed another part of the challenge that would become important across the competition: understanding a problem deeply does not automatically create a pathway to deployment.

When judges asked how farmers would discover and adopt the system, who would build it beyond the prototype stage and how a free tool could remain financially sustainable, the team was still developing those answers. What they understood clearly, from their own family experiences, was why preserving this knowledge could matter to farmers and why they wanted to solve the problem.

That did not make the underlying idea less valuable. It showed what Land Memory AI would need next. The team had identified a form of agricultural knowledge worth preserving and designed an unusually approachable and simple way to capture it. The next challenge was figuring out how to turn that strong understanding of the problem into something farmers could reliably access, adopt and continue using.

Jessica Parra Serena and Thi Xuan My Tran of East Texas A&M took a deliberately direct approach to one of agriculture’s most basic decisions: when to water and how much water to use.

Their system, Irriga, combined in-ground soil and well sensors with weather information to give farmers an irrigation recommendation and explain why. It was also designed to work online or offline, an important consideration for farms where reliable connectivity cannot be assumed.

The simplicity of that approach immediately stood out to me during judging. If the question is how much moisture is actually in the soil, there is an intuitive appeal to putting a sensor in the ground rather than trying to infer everything indirectly from satellites or aerial imagery.

Irriga was my initial choice for the winner.

But another judge asked a question that exposed how quickly a simple idea can become a deployment problem: How many sensors would a farmer actually need, and how far apart would they need to be?

The team did not yet have that answer. And it mattered. A system that works with a few sensors in a prototype may look very different when spread across hundreds or thousands of acres. Sensor density could affect cost, installation, maintenance, scalability and whether the readings accurately represent conditions across different parts of a farm.

That question did not undermine the logic behind Irriga. It identified the next technical problem the team would need to solve before the simplicity of the prototype could translate to farm-scale deployment.

The judges also pressed on what would happen beyond the prototype: who would build the complete technical system, how farmers would gain access to it and what would be required to bring it into broader use. The team was still working through those questions.

Only after judging ended did I learn an important piece of context. In a conversation with Parra Serena, she explained that teams had not been required to develop a formal go-to-market plan. Irriga’s team had approached the competition primarily from a business rather than technical background, so they had focused on the problem and the concept rather than building out a detailed technical deployment strategy.

That information changed how I looked back at the judging. It did not answer the sensor-spacing question, which remained a real feasibility issue. But it did make me think differently about how heavily some of the implementation questions should weigh when a formal commercialization plan had never been part of the assignment.

At the same time, all five finalists had presented solutions to real problems that could potentially help real people. We still had to find meaningful differences among them and ultimately rank the projects, which meant questions about technical feasibility, deployment and what would happen next became part of that separation. Irriga did not ultimately win, but its direct approach to monitoring and measurement still made the most intuitive sense to me: if you need to know what is happening in the soil, measure the soil.

Irriga’s core idea remained compelling because it was so direct: measure the conditions farmers actually need to understand, combine those measurements with weather information, and turn the result into a clear recommendation about when and how much to irrigate. That could help farmers avoid watering when rain is already on the way or, when water needs to be conserved, determine whether existing soil moisture means irrigation can safely wait another day or two.

The harder question was what it would take to make that simplicity work at the scale of a real farm.

Texas A&M University-Corpus Christi graduate students Maharshi Barot and Saumya Patel formed team CoderOP around a deceptively basic water problem: utilities can spend money treating water only to lose some of it through leaks before it ever reaches customers.

Their system combined nighttime water-flow data with free satellite imagery to identify areas where leaks were more likely to be occurring. But instead of handing utility workers another dashboard full of AI analysis, CoderOP was designed to turn that information into something much more practical: a plain-language worklist showing which areas may deserve attention first, where they are located and how serious the problem may be.

That immediately made sense to me during judging.

I remember telling the team that I could not believe more of this process was not already automated. The team described a largely manual workflow in which utilities may receive large numbers of alerts and pressure changes that someone still has to sort through individually. It can take days to work through those signals, and the students said a potential leak may sometimes wait a week or longer before someone is available to investigate it.

That delay can grow even longer if the first visit only confirms that a leak exists. A worker may then need to determine what repair is required, obtain or order a part and return to complete the work.

CoderOP was designed to shorten the earlier part of that chain. If pressure drops or another anomaly appears, AI can help evaluate whether the signal looks like a false positive or something that warrants human attention, estimate its severity and give the utility worker more precise information about where to investigate.

The goal was not for AI to declare that a pipe was leaking and send someone to dig it up. It was to help a limited utility crew decide which possible problem deserved its attention first — and arrive with better information when someone did go out.

That distinction matters for smaller or under-resourced utilities. Field crews cannot investigate every unusual reading at once, and false positives can consume time that could be spent on genuine leaks. CoderOP offered a way to use AI as a filtering and prioritization layer before a human crew went into the field.

The potential water savings also gave those decisions urgency. The team told judges that even a relatively small leak could waste as much as 1,300 gallons of water per day. If a leak remains unresolved for days or weeks, those losses can accumulate quickly — particularly in communities already dealing with water scarcity.

CoderOP therefore approached AI somewhat differently from Irriga. Irriga was designed to help a farmer decide when and how much water to use. CoderOP helped utility workers identify, understand and prioritize where water might already be escaping.

During judging, I personally gave somewhat more weight to systems where the AI was helping make a direct resource-allocation decision, such as when to irrigate or where scarce water should go. That was not a judgment that CoderOP’s technology was less capable or its problem less important. It was one of the distinctions I used because all five projects were strong enough that we had to find ways to separate them.

And CoderOP solved an important part of the water problem in a particularly accessible way. Utility workers did not need to become data scientists or manually work through every alert before deciding what deserved attention. The technical analysis happened behind the scenes, while the person responsible for the next step received something much simpler: a prioritized list showing where to look, how urgent the problem appeared to be and which alerts might not require a field response at all.

Once again, the AI did the complex work in the background so the person on the other end could focus on the decision that mattered.

For twin brothers Edgar and Eduardo Bello Gonzalez of the University of Texas Rio Grande Valley, drought was not an abstract environmental problem they encountered for the first time in a hackathon prompt.

The brothers grew up in Mexico, where they experienced recurring drought, water shortages, crop losses, impacts on livestock and increased fire risk. After moving to Texas, they encountered many of the same pressures in the Rio Grande Valley. Edgar has said the problem had been on their minds for roughly four years before more recent work with mapping drones helped them recognize how the technology might be applied.

That experience became Dos Ojos — “two eyes.”

The first eye was free Sentinel-2 satellite imagery, which could provide a broad view of crop conditions across a farm. When that wider view flagged an area of concern, the second eye — a low-cost camera drone — could examine the area more closely. The drone imagery could be used to reconstruct a 3D view of the crop canopy, while computer vision helped measure canopy volume, identify dead or missing plants and map stress row by row.

The system then combined those perspectives to help determine where limited irrigation water should go.

But farmers were not expected to interpret satellite imagery, drone maps, computer-vision output or a complicated dashboard themselves.

The complexity stayed in the background. Dos Ojos could give farmers a detailed picture of crop conditions, explain in everyday language what the system was seeing and why certain areas appeared stressed, and highlight what deserved closer attention. That analysis could then lead to a clear recommendation — irrigate, watch or sacrifice — with the explanation and recommendation delivered through a short bilingual text message while leaving the final decision with the farmer.

The word “sacrifice” sounds severe because the decision itself can be severe. When water is heavily rationed, a farmer may not have enough to protect every field. Spreading limited water across everything could leave a larger portion of the crop unable to survive. Dos Ojos was designed to help identify where available water could still make the greatest difference and, in an extreme situation, where a farmer may need to stop allocating scarce water.

That made the AI’s role particularly consequential to me during judging. It was not simply identifying that crops appeared stressed. It was helping turn limited information and limited water into a recommendation about where a scarce resource should go next.

The brothers had also already discovered where their own user experience needed work. To use the system, farmers had to identify their property in Google Maps and draw a boundary around the farm so the satellite analysis knew where to look. The brothers told judges that farmer feedback had identified this as the most difficult part of the process. Simplifying that step was one of the improvements they planned to work on next.

That feedback mattered because Dos Ojos was being designed around farmers who might not want to navigate complicated technology simply to obtain an irrigation recommendation. The goal was not only to make the analysis sophisticated enough to be useful, but to keep reducing the amount of technical work required from the farmer.

The team applied the same thinking to cost. The brothers presented a three-tier pricing model designed to give farmers access at different price points. A lower-cost option would rely on satellite information alone. A middle tier would add drone imagery, while a higher tier would add thermal imaging to provide a more detailed view of crop conditions.

That gave farmers a way to choose the level of information they could afford rather than making the most expensive version of the technology the only way to participate. In a region where drought can already put financial pressure on agricultural producers, that accessibility became another important part of the project’s design.

The problem itself was very real in the region the brothers were designing for. Severe water shortages in the Rio Grande Valley have already caused significant agricultural losses; in 2025, the U.S. Department of Agriculture announced a $280 million agreement with the Texas Department of Agriculture to provide economic relief to eligible regional producers affected by the shortage. USDA said water shortages had already ended regional sugarcane production and threatened crops including citrus and cotton.

Then, shortly before the final judging, something changed for Dos Ojos.

The brothers told us that one of them received a message from a professor offering to support the project and help them pursue USDA funding. About ten minutes later, the other brother received a similar message from another professor. Within minutes, their university dean had also contacted them offering support and help pursuing funding.

They had not secured a USDA grant, and no specific federal funding was guaranteed. What they suddenly had was something different: two professors and university leadership willing to help them figure out what came next.

That mattered.

Several teams had arrived with strong ideas but still faced questions about who would continue development, where resources would come from or how a prototype could move toward deployment. Dos Ojos could now point to people inside its university willing to support the project and help the brothers explore potential funding opportunities.

The project had moved, at least one step, from “What could this become?” toward “Who can help us keep building it?”

That did not make Dos Ojos perfect. The farmers’ difficulty with the mapping step showed there was still usability work ahead. But the brothers could identify the problem, explain what they intended to improve and point to people willing to help them continue the work.

When the judges were trying to distinguish among five strong finalists, that larger combination became difficult to ignore: lived experience with the problem, a technically credible satellite-and-drone system, farmer feedback, an accessible delivery method, pricing designed around different budgets and newly emerging institutional support.

Dos Ojos gave us one of the clearest answers to the question that had come to define the judging: not just whether the idea was good, but whether there was a believable path for the idea to keep moving after the competition ended. After meeting Edgar and Eduardo, it was difficult to imagine the project ending with the hackathon.

By the end of the presentations, the judges had a problem we had not solved by asking more questions.

All five projects were worth continuing.

That did not mean they were identical or equally developed. Each had different strengths and different questions still ahead of it. But there was no project we could simply dismiss as a weak idea or an AI demo looking for a problem.

And we still had to choose a winner.

That was when the questions we had been asking throughout the day took on more weight. Sensor spacing mattered because Irriga eventually had to work across a real farm. Adoption and sustainability mattered because Land Memory AI needed a way to reach the farmers it was designed to help. Data access and cybersecurity strengthened NoNiMo because the team could already answer some of the questions about what continued development would require. CoderOP showed a clear operational use for AI, while Dos Ojos had begun assembling not only the technology but some of the people and resources that could help move it forward.

My own thinking had shifted during the judging. Irriga had initially made the most intuitive sense to me because of its direct approach to measuring soil conditions. NoNiMo impressed me with what the team had managed to build in only a few days and how much technical thinking was already behind it. And as we learned more about Dos Ojos — the farmer feedback, pricing structure, simple delivery, and the professors and university leadership beginning to rally around the project — its path beyond the hackathon became increasingly tangible.

That was ultimately what made the decision so difficult. We were not trying to find the one project with value and four without it. We were trying to decide which combination of problem understanding, technical execution, usability, resources and readiness gave one project an edge on that day.

Before the winner was announced, each judge was asked to offer a few closing thoughts to the students.

By then, I had spent the day looking for weaknesses in their projects because that was part of my job as a judge. But after questioning them, comparing them and trying to find enough separation to rank them, I had come away believing something very different about the group as a whole.

So I told them:

“Even if your team doesn’t win, you have to keep pursuing these projects, okay? Each of your projects is a real solution to a real problem. I hope you keep pursuing your projects because your community needs you, your society needs you to bring this solution to market. You have to keep going even if you don’t know where this is leading. Your solutions matter.”

Four of those teams were about to leave without the grand prize.

I did not believe any of them should leave believing their project had failed or that it shouldn’t exist.

And that feeling extended beyond my own closing remarks. All of us judges came away from the competition more optimistic about the next generation of problem solvers and about what AI could look like in their hands. These students had started with needs they saw in their own communities and spent their time trying to build something that could genuinely help.

Whatever the final ranking, that was difficult to walk away from without feeling a little more hopeful about what comes next.

Then the winner was announced.

Dos Ojos had won the $15,000 grand prize.

For Edgar and Eduardo Bello Gonzalez, the moment was emotional. Edgar later described the win as one of the best experiences of his life. Eduardo wrote that the competition had given them a reason to begin a project that “grew more than expected,” and said they intended to continue developing Dos Ojos and expanding its impact.

The prize gave them another resource to do that. Combined with the professors and university leadership who had already offered support and help pursuing funding, the brothers left the competition with more than recognition. They had people willing to help, money that could support the next stage of development and a clearer sense of where the system still needed work.

But the most important result of From the Ground Up may not have been that one team won.

Across the five finalists, students had built systems to help planners decide which drought scenarios deserved deeper modeling, help utilities identify where treated water might be leaking, preserve generations of agricultural knowledge, tell farmers when and how much to irrigate, and help determine where scarce water could do the most good.

None of those systems required the person using them to become an AI expert.

A hydrologist could receive a shorter list of scenarios worth examining. A utility worker could get a prioritized worklist instead of sorting manually through overwhelming alerts. A farmer could preserve years of experience through photos, voice notes and a familiar scrolling feed. Another could receive an irrigation recommendation based on what was actually happening in the soil. And Dos Ojos could translate satellite imagery, drone analysis and computer vision into an explanation of what was happening in a field and what the farmer might want to do next.

That may be the clearest lesson from From the Ground Up: useful AI does not have to announce how sophisticated it is. It has to help someone make a better decision.

The technology mattered. But so did knowing who was going to use it, what information that person actually needed, what resources they had available and what decision was waiting on the other side.

That is what these five teams understood unusually well.

And in that sense, the hackathon did what Zoho had set out to test. The students were close enough to these problems to understand why they mattered, see how they affected people around them and recognize where technology might genuinely help. Their connection to the problem gave the projects urgency, but it also gave the students something more important: a reason to keep working on them.

Five teams of college students looked at problems their communities were living with and decided they could do something about them. They did not wait for someone with more experience, more authority or a bigger title to decide the problem was worth solving first. They understood these challenges from close enough range to recognize where technology might genuinely help, and they used the tools and skills available to them to start building.

The institutions around them still mattered. Zoho created the opportunity, and in some cases colleges, professors, experts and university leadership helped open a path forward. But the first step came from the students themselves: they saw a need in their communities and decided they were capable of contributing to the solution.

They did not begin with AI and search for somewhere to apply it.

They began with problems their communities were already living with and asked how AI could help.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing support, AEO/GEO/SEO optimization, image concept development, and editorial structuring support from ChatGPT, an AI assistant. All final editorial decisions, perspectives, and publishing choices were made by Alicia Shapiro.