The AI engineer quietly building Pakistan's food future

Computer science chose him — he didn’t choose it

The episode opens with Muzamil introducing Atique Ur Rehman as a batchmate from FAST Islamabad’s computer science programme — someone he describes as “one of the smartest guys you’ll ever hear” across two very different domains: technology and organic farming.

Atique grew up in Wah Cantt, a small city attached to Taxila near Islamabad, built around the Pakistan Ordnance Factory. He was a strong student in government schools, good at maths and physics, and fully expected to become an electrical engineer. He had sat the entry tests for NUST and UET Lahore when the test was leaked and cancelled. In the gap that followed, a friend mentioned FAST and a scholarship that was restarting after four years.

“It chose me. I didn’t choose computer science,” Atique says. The ICT R&D scholarship had a restriction that applicants could not be from a developed city — which, for once, being from Wah Cantt was an advantage. He sat the test, was selected the same day, and was still planning to wait for UET to reconduct its exam when a friend at FAST told him to attend classes for two or three weeks. Within three weeks, he was sold.

The scholarship itself was remarkable in scope: full fees, pocket money, books, residence, and transportation. Atique is emphatic that it is “a very underrated program, and it’s not appreciated enough.” The NICT model trained government school teachers first, who then delivered coaching to students in their own areas — making entry-test preparation free and accessible to children from Dadu and Hazro who had never attended a coaching centre. The programme ended after a few years, but the lives it changed are visible today.

Why communication is the real bottleneck for Pakistani engineers

Muzamil presses Atique on something that is easy to overlook: the quality of his communication. He is from a government-school background, from a comparatively smaller city, and came through FSC rather than O-levels — all the stereotypes that would predict poor English and weak professional communication. Yet the conversation is fluid, precise, and self-aware.

Atique traces it to two things: FAST’s deliberate effort to build communication skills, and early exposure to international clients. His first company had people from fifteen nationalities. He had to learn not just English but how to communicate without offending, how to understand what a client was actually asking beneath the surface of their question, and how to navigate work-culture differences — including something as specific as explaining to German colleagues why he needed to leave a meeting for prayer.

“If you’re only talking, that’s not communicating,” he says. “You have to understand what the person is asking. Sometimes the question is not in the exact question — the question is inside.”

Muzamil makes the point directly: Pakistani universities are not doing enough on work ethics and personality development. Atique agrees. “As an engineer, I think even at early stages, 50% of your work is actually technical, and the rest is communicating.” Engineers who cannot communicate their problems cannot get them solved. Engineers who cannot read a client’s requirements cannot build the right thing. The technical curriculum is improving; the human side is not keeping pace.

Building AI before it was a buzzword

Atique’s final year project at FAST — done with Muzamil’s brother Mohsen, who was working in esports in Silicon Valley at the time — was a location-based augmented reality app that let users attach memories to physical places and see others’ notes when they pointed their camera at the same spot. The project was called Snapchat. It introduced him to computer vision and convinced him to do a master’s instead of taking a job.

During his master’s, he and a small group of students and faculty founded REVEAL Lab — Recognition, Vision and Learning Lab — at FAST. AI was not yet common in Pakistani universities. They ran a five-day workshop for teachers from eighteen to twenty universities, funded by HEC. They applied to Nvidia for a GPU when FAST had none, and Nvidia gave them a K80. “It was a very memorable moment for me at least,” Atique says.

From there, a German startup — Automotive Artificial Intelligence, founded by former Audi engineers — came to the lab looking for talent. Atique joined as technical lead for Pakistan. When he left two years later, the Pakistan office had grown from four or five people to 120. The core AI work was happening in Islamabad, not Berlin. “They were doing cutting-edge work. The whole core AI was happening here in Pakistan.”

What AI is actually doing now — and what it is not

Muzamil asks the direct question: is AI a buzzword or is it real? Atique’s answer is structured and honest. The first phase was excitement without substance — companies attached AI to their names to get funded, engineers attached it to their CVs to get hired, and neither side knew what they were actually looking for. Real-world deployment was a brutal check: models trained in lab conditions fell apart on live data.

The second phase is different. “Companies have started to realize what type of engineer they need.” At Motive, where Atique now works, the AI system processes around 100,000 videos a month from dashcams on commercial vehicles, identifying dangerous driving events — close following, phone use — and alerting fleet managers in near real time. “While we are sitting here, 20,000 of those possible incidents might have stopped.” Human annotators were tried and abandoned; the scale is simply not possible without the model.

He pushes back on the job-destruction narrative. “AI is not taking your job. It’s giving you another job with more meaningful impact.” The 500 people who would have had to watch truck-driver footage all day are instead doing work that requires human judgment. The jobs that AI is replacing, he argues, are jobs that no human mind should be doing in the first place.

On where AI goes next, Atique points to two areas: medical imaging, partly because of his own experience of being misdiagnosed, and natural language interfaces — the ability to talk to software in Urdu and have it do tasks that previously required learning complex tools. He describes work happening in Pakistani labs on Urdu conversational agents, and agrees with Muzamil’s framing that the real value is in making digital products inclusive for the working class — the majority of the economy — not just the educated middle class.

Two months believing he had two years to live

Later in the discussion, Atique describes a period that reordered everything. Around 2018, he went for a routine medical check and a physician noticed a mass on the side of his chest. After an ultrasound and CT scan, he collected the report without the doctor present, read the diagnosis — left wing sarcoma — and Googled it. The first result said maximum two years to live. He fainted.

He was engaged at the time. His family was planning the wedding. He told no one for two months while oncologists continued their diagnosis. “I couldn’t tell my family because I’m very attached to my father and mother. And I knew it would kill them.” His support system was his friends, who accompanied him to appointments and let him stay at their homes while his mother was choosing wedding dresses.

The final diagnosis was fibromatosis — an aggressive but non-cancerous tumour. Surgery followed, removing part of the mass and some ribs. A year later it began growing again, requiring a second operation. By the time of the second surgery, Atique was managing farm construction from the hospital, recently married, and sitting through Motive’s seven-round interview process from his hospital bed.

What the experience gave him, he says, was the ability to handle stress — and a conviction that mental health care is not a taboo. A friend’s brother, a psychologist trained in Germany, helped him break the loop of catastrophic thinking. “The problem is that when you’re overthinking, you’re not able to discriminate between the possibility and something that is not possible.” He still visits the psychologist regularly, for productivity, work-life balance, and ordinary life decisions. “We should practice this more.”

What he saw in the livestock markets

Muzamil frames the pivot to farming as the most counterintuitive part of the story: a deeply technical person, voluntarily walking into one of the most stereotypically rural industries in Pakistan. Atique explains that it started personally — he wanted to raise his own Qurbani animal to preserve the actual sense of sacrifice, rather than buying expensive meat at double price with no connection to the animal. He rented space near Wah Cantt, partnered with someone local to care for the animals, and did this for a couple of years.

Then he started visiting livestock markets to buy animals, and what he saw changed the direction of the project entirely.

“I started to realize that what kind of meat are we eating?” The markets were dominated by two categories: animals slaughtered early because of disease or injury, and animals that looked healthy but had been raised on steroids — hormones like TEGSA and PPA, originally designed to support liver function, given weekly until the liver stops functioning on its own. “The math doesn’t add up” for a healthy, naturally raised animal to reach 80 or 90 kilograms. When the numbers don’t add up, something else is happening.

He extends the analysis to poultry — broiler chickens reaching one and a half kilograms in twenty-one days, kept alive through heavy antibiotic use, with no regulatory check on excretion periods before slaughter. To dairy — the hormone RBST used to extend lactation for years beyond the natural cycle. To produce — a friend’s goats wandered into a neighbouring maize field for five minutes and all of them died from pesticide exposure. “That’s the same maize that’s going to your table.”

Pakistan, he notes, now has the highest percentage of its population on some stage of diabetes of any country in the world — thirty-one percent. He connects this directly to what is in the food supply.

Breeding a uniform goat from scratch

Maveshi Farms began as a Qurbani solution for city dwellers — buy an animal a month or two before Eid, visit it, receive updates, have it delivered and professionally slaughtered. That was phase one. Phase two is more ambitious: developing a locally adapted, genetically uniform breed of goat that can perform predictably on natural feed, so that the model can be replicated by other farmers.

The problem Atique identified is that Pakistan has no pedigree system for livestock. You can go to a market and buy an animal that looks good, but you have no idea what characteristics it will pass to its offspring. International breeds like the Boer goat come with documented performance data — feed conversion rates, weight gain per day — because they have been selectively bred for uniformity over generations. Pakistani breeds have not.

“We tried to find good male bucks and then we did our research. We had to go to Bawal, go to houses — where did this animal come from? Where did its father come from?” Atique and his partners have assembled a collective pool of around 250 to 300 goats across several farms, working with a strain of the Beetal breed called Makigina. They are tracking which animals show strong immunity, early weight gain, and low health problems, and pairing them selectively. A good sire can cost ten to fifteen lakh rupees. They are paying it.

The long-term vision is a replicable model farm: someone buys a herd from Maveshi Farms, follows documented practices, and gets predictable results. Right now, Atique says, people invest in thirty or forty animals, have no idea what to feed them, and are out within six months. “There is no documentation. Nobody knows what to feed them.” He is working with a mentor in the US who is liaising with Penn State University nutritionists to develop a feed formula built from locally grown inputs — reducing import dependence while cutting costs.

By the end of the conversation, Muzamil puts it plainly: “I hope that’s your billion dollar unicorn.” The man who built AI infrastructure for self-driving cars is now trying to build the genetic infrastructure for Pakistan’s livestock industry — one carefully selected buck at a time.