It was 2019, and with the boundless optimism of fools, we embarked on creating a flexible-bodied Unmanned Ground Vehicle for the noble pursuit of agriculture. Then, as if orchestrated by a particularly malevolent deity, COVID-19 descended, and with everyone locked down tighter than a duck's proverbial, we weren't quite sure what to do with our shiny new toy. Its brain, an ostentatiously expensive NVIDIA Jetson AGX Xavier, seemed far too clever to be merely shackled to a chassis. So, marooned at home, observing farmers wrestling with the delightful combination of lockdowns and a vanishing workforce, a thought struck my beleaguered mind: why not estimate the yield of grapes, oranges, apples, and other such tedious baubles? Better workforce management, you see.

This very isolation, much like the theme of this entire wretched saga, led to me ranting about the utter unimportance of AgTech in Türkiye on LinkedIn around 2020. Lo and behold, some chap named Serdar Dikbaş stumbled upon my digital diatribe and, clearly a glutton for punishment, wanted to meet – online, of course, given the global house arrest. Several phone calls ensued, during which he revealed, "I'm thinking about starting a series called Agricultural Technology. Would you be able to explain what you're doing?" "Naturally," I replied, with the feigned modesty of a peacock. In the brave new Skype-connected world, we ended up on an actual TV program, aired on AgroTV. During this momentous broadcast, Serdar, with touching earnestness, declared, "Very promising technologies indeed, hope to see them perform in real life as well." To which I, still battle-hardened from the endless, futile skirmishes between AgTech and Defence funding, retorted, "Just give me coordinates. We will be there." And I bloody well meant it.

A week later, the coordinates arrived. I promptly grabbed my desktop PC, laptop, a monitor, two drones (one for NIR, one for RGB, naturally), several action cameras, and deployed the entire YieldEstimator arsenal. This included the aforementioned fancy Jetson Xavier and relics from my previous groundbreaking folly, Spectralix – a system from 2016 that performed real-time NDVI calculations using deep learning when most in Türkiye thought NDVI was a new boy band and "deep learning" was something you did in a particularly boring lecture. Back then, I’d even cobbled together a tablet and drone camera that did this NDVI wizardry on the fly using a Jetson TX1, later a TX2. But I digress. On site, with no internet to speak of, I collected data, labelled it in the field like some digital peasant, deployed it on the NVIDIA Jetson AGX Xavier, connected a Logitech C922 webcam, strapped the contraption to the tractor's windscreen, and hit play. And by Jove, it did. A staggering 74% accuracy on its first-ever real-life demo, on "walnuts" of all things. Meanwhile, as others munched on their meatball sandwiches, I, in a fit of pique, launched a drone, did a quick flight, landed, extracted frames, labelled data, and deployed another model on the Jetson. This one achieved 100% accuracy in detecting trees. Everyone present was, to use the vernacular, awestruck. Or possibly just bewildered. This minor triumph led to Serdar inviting me back to the AgroTV studios for two more programs, where we waxed lyrical about UAVs, UGVs, and Deep Learning in Agriculture. Oh, the heady days.

2020 itself was consumed by the glamorous task of downloading over a thousand videos of fruit from the darkest corners of the internet. Frames were extracted, single-handedly labelled (a truly Sisyphean endeavour), and models were trained. Simultaneously, I frantically sought a model, a method, an algorithm – anything, really – to somehow count the damn things. After a dash of tinkering and a metric tonne of hardware and software wrangling, I'd actually cobbled together something that, against all odds, worked!

Our first paying victim, Codorníu Raventós in Spain, signed up for a three-phase circus. Alas, the driver of our van had an "unfortunate event," cutting that particular farce short. Still, that baptism by fire with Edge Processing and action cameras taught us a lot. Them too, presumably – like not to trust enthusiasts with their vehicles. That year, we'd achieved the impossible: real-time yield estimation on the fly with Edge AI, whilst careening through vineyards at 20kph in a van bristling with action cameras filming at 240fps, all to count bloody grape bunches.

SeeTree from Israel, bless their optimistic hearts, wanted an exclusive licensing deal for TARSENS' YieldEstimator. We declined, naturally, perhaps a touch too haughtily. Then, just as we were wooing some Indian Kiwifruit magnates for a cool $15M, Russia decided Ukraine looked tasty. Investors, predictably, shat their collective trousers. Fifteen million dollars vanished like a fart in the wind. The team, understandably miffed, disassembled with profound disappointment.

Come 2022, an Italian ray of sunshine named Gabriele from Finest Grape Tech appeared. He actually helped, procuring the specified action camera, laptop, and a sacrificial vehicle for vineyard jaunts. He sent footage; we processed and mapped many a vineyard, feeling briefly competent.

Then came January 25th, 2023. We found ourselves in Adana and Hatay, Türkiye, for more in-field demonstrations. We met with producers of pomegranates, apples, bananas (yes, bananas in Türkiye, don't ask), and citrus. One particularly accommodating citrus producer, thanks to Serdar Dikbaş's enviable connections, willingly halted his entire harvest for our little experiment. We, of course, had zero prior experience with citrus, but I was keen to see what the system could make of data from only one side of the mandarin trees. Our final estimation clocked in at 50% of the final weight which, given the circumstances, I chalked up as a resounding success. It was during this illuminating test that I also discovered low-hanging powerlines can royally mess with a GoPro's internal GPS, forcing us to invent a new mapping method on the spot. We had a rather enjoyable time, all things considered. Ten days later, Türkiye's biggest twin earthquakes ripped that very land apart, claiming over 40,000 lives and, incidentally, any chance of us working there again. May the dead rest in peace; our prospects certainly did.

Throughout this rollercoaster of despair, we were occasionally invited to give talks. We were even shortlisted as Vision Tank startup finalists for the Embedded Vision Summit in Santa Clara – Silicon Valley, for those unacquainted with geographical trivia. But, of course, I didn't have a US visa. So, we couldn't attend in 2022. However, a delightful clause in their agreement stated that in such predicaments, one could attend the following year free of charge. Heartened, I applied again in 2023, roped in some long-suffering friends to display our videos on a TV (thank you, you sainted fools!), and basked in the fleeting glory of seeing my startup's name alongside giants like Sony and Arm. Such great joy. Zero actual return. Another time, the New South Wales Department of Primary Industries in Australia expressed keen interest. We presented our tech to them, utterly and extremely nakedly. I told them, plainly: "If we can't see them, we can't count them. That is why it is called Yield-Estimator; the estimation comes from approximation." These approximation methodologies were our own cunning inventions: Enhanced Parallax Correction (EPC), Occlusion Correction Factors (OCF), and the art of spotting 'Unreal bunches'. Don't ask for details; they're a trade secret, buried with the company.

And the vineyards! A very well-known one in Türkiye, whose founder sagely advised me to package this marvel as simple PC software for the "average Joe." I retorted, "Even Microsoft hasn't managed to protect its software for over 30 years. How on earth am I supposed to protect mine?" Crickets, as they say. Then there was a US vineyard, which demoed a pseudo-competitor, Bloomfield, with their 'Flash' setup, funded by the Washington State Wine Commission. The result? "Ultimately, results were disappointing," they reported. And what happened next? Bloomfield partnered with Monarch Tractors, which promptly got investment from Kubota. Meanwhile, our humble tech boasted 99.778% counting accuracy in Napa Valley and 87% weight accuracy in Italy – both vineyards we'd never physically visited, mind you, just processed their action camera videos. The injustice is almost poetic.

Damn it, I simply must mention academia and research institutions. MAN! THEY ARE INCOMPETENT! Did you get that? No? Let me repeat: THEY! ARE! IN-COM-PE-TENT! I have never encountered such a concentrated mass of dull, empty, repetitively insistent, stupid people. Not in one place, thankfully; they're distributed, like a particularly virulent plague. Let's start with Australia. We were offering a free remote demo of a row or two for those with compatible GoPros (Hero 9-10-11). Some Aussies scanned two rows, sent the data, we processed it, sent back results. They liked it! So much so, they wanted it "again" and "for free." I said NO. In France, a "guru" of wine producers, lauded as the Einstein of viticulture, followed our every move, sniffing for technological breadcrumbs. He even had the audacity to tell me, "Without my help, you won't survive in this space." Well, we're dead, but it doesn't mean he was right; it means they still don't have the tech, even with their pseudo-Einstein. Then there was one from the United States, a compatriot no less, who tried harder than anyone I've ever met to make my system look incompetent. He intentionally held the camera incorrectly, deliberately misspelled my company name on social media, and wilfully ignored our system's success in his own vineyards. And the final straw, from Türkiye: a charlatan who tried his level best to extract keywords from my technologies to cobble together a paper, paving his cockamamie path to professorship while trashing a million-dollar technology for a few meaningless points on a publication no one will ever read. To all such individuals: this technology is dead. I built it. I killed it. It has no patent, no white paper, no manuscript. It was a secret know-how, triple-encrypted on an SSD chip, which was then hard-desoldered from its motherboard, sealed in a vacuum bag, and now resides in a bank vault. Want it? I told you the price. May 30th, 2025, is the final deadline. After that, it becomes legend.

2023: we officially launched YieldEstimator Inc. Three weeks later, Silicon Valley Bank imploded – a fitting overture to our grand opening. We had precisely zero customers, partly due to my gammy leg and mostly because my co-founder was about as useful as a chocolate teapot, especially at marketing. Oh, and the fraudster forgot to mention Delaware franchise taxes. I informed him I was buggering off to do a PhD in physics and had no time for this charade: either he paid the taxes and shuttered the company, or I would. The coward didn't hesitate for a second, offloading all his shares onto me and scarpering like a rat up a drainpipe. Leaving me, naturally, to foot the bill for all the taxes and hire an agency to perform the corporate euthanasia.

In 2024, EIT, in a moment of what I can only assume was temporary insanity, sponsored a test farms demo in Spain with Bodegas Vetusta. We had a jolly time. For starters, they didn't possess a single action camera with GPS. So, I dispatched mine via DHL, who, in their infinite wisdom, managed to shaft me on both the sending and the (attempted) return journey. They eventually attached the camera, captured the data, sent it over, and we processed it, delivering the KML files. We even unearthed fascinating insights like animal damage and slope-to-yield distribution – pearls before swine, as it turned out.

All told, we'd flaunted our wares in Canada, the US, France, Spain, Portugal, Italy, Germany, Turkey, Australia, and New Zealand. Connections globally, you see. We were very important. I genuinely believed this venture would become a cornerstone of wine production. I even applied to join Şarapder (the Turkish Wine Producers Association). They declined, naturally. Just like every other sodding farmer.

In my now razor-sharp, if somewhat jaded, opinion, farmers don't want tech. No matter how affordable, how bleeding simple you make it, they'd rather stick to their archaic methods and literally count their own damn grapes or kiwis, probably using their toes.

Let me regale you with two final anecdotes that perfectly encapsulate this tragicomedy:

In a Napa Valley vineyard, some hapless intern filmed with an iPhone SE 2020. We counted 1301 bunches. They reported back: the intern had manually tallied 2870. I was, shall we say, perplexed, given our supposed 90% accuracy. "Which row did he count? Both sides?" I queried. "Dunno," came the enlightening reply from the viticulturists. They dispatched the poor lad again. This time: 1356. I disabled our 'safety' parameters (the ones designed to prevent us looking too clever) and our system spat out 1353. He'd wasted two hours; our video was three minutes long. Progress, eh?

Then there was the Californian kiwifruit plantation. They strapped a GoPro to an ATV, complete with floodlights (because, why not?). We processed the videos and found something rather curious: the yield on the right side of the orchard was triple that of the left. A quick peek at the neighbours revealed an Angus farm conveniently located next to the bountiful right side. Coincidence? I think not.

In both instances, did the farmers want to verify our potentially revolutionary findings, either privately or publicly? Did they hell.

Why do I bang on about it this much? Because I cared. I poured my soul into this. You, the average Joe farmer, were so damn incompetent, so breathtakingly shortsighted, you couldn't grasp how advanced it was. I know you can't imagine it, but try, just for a minute. You buy a commercial, off-the-shelf action camera (GoPro Hero 9, 10, 11, even the 13 for God's sake!), attach it to your vehicle, drive through your vineyards at up to 20kph, send the raw data to us, and we geolocate your grape bunches, pinpoint diseases, and estimate the weight of your yield without even leaving our office. Instead, you chose to pay some charlatan 'advisor' to amble through two or three rows, bill you a couple of thousand dollars, and give you an estimate that's 50% accurate at best. Your total expense for YieldEstimator would likely never have even hit $999. Live with that delightful fact for the rest of your life.

And that, dear reader, is why I say farewell, YieldEstimator. I gave it my all. Unfortunately, it seems I predominantly encountered the agricultural equivalents of Particularly Stupid People. I now have a profound, almost painful, understanding of what Nikola Tesla must have felt. A monumental, soul-crushing disappointment.