The Ice Sculpture Principle: How AI Engineers Build What Must Not Break

An Interview with Machine Learning Engineer Kriti Goyal

Two Days Behind

Imagine you’ve missed two days of work. Not a week. Not a month. Just two days.

When you come back, the world has shifted. A new model has dropped. A breakthrough paper is everywhere. A tool you relied on has already been replaced by something faster, cheaper, and better.

For Kriti Goyal, this isn’t an exaggeration—it’s her daily reality as a machine learning engineer. “The pace is overwhelming,” she says. “If you step away for even 48 hours, it feels like you’ve missed so much.”

And yet, she speaks about it with something closer to awe than exhaustion. For her, this relentless speed isn’t just a challenge. It’s what makes working in AI exhilarating. It’s the reason she’s fascinated by the field’s rapid democratization: small, efficient models that no longer demand a supercomputer or a massive cloud bill to run.

“Today, a hospital can deploy its own model locally,” she explains, “and take care of its own privacy concerns without sending sensitive data anywhere else.” In her mind, this is more than a technical shift—it’s a quiet revolution, one that is pulling AI out of centralized labs and placing it directly into the hands of those who need it most.

The Ice Sculpture Principle

Kriti has a favorite metaphor for her work: “AI is like an ice sculpture—it has to hold under pressure and work flawlessly.”

She laughs when she says it, but she’s not joking. It’s her way of describing the unforgiving nature of production AI. Demos are easy; anyone can hack together a proof of concept that works in ideal conditions. But real-world systems? Those are built to withstand heat lamps, impatient users, and the occasional hammer someone swings at your creation just to see if it will break.

“Building a demo is one thing,” she says. “But turning that into something reliable, something that people can actually trust, is a completely different challenge.”

That mindset—part precision engineering, part entrepreneurial grit—shapes how she sees the future of AI: tools that are not only powerful, but robust, private, and safe enough to be trusted in mission-critical environments.

From CV to NLP: A Personal Journey

Kriti didn’t set out to become a machine learning engineer. Her path started in computer vision, pulled in by professors whose enthusiasm for the field was contagious. “They showed me how much you could do with images and algorithms,” she recalls. “It was like magic, but grounded in math.”

Back in 2019, much of her work was rooted in traditional machine learning. “We were still in this world of feature engineering and heuristics,” she says. “Rules were easier to understand.” But as deep learning swept in, everything changed. The models became larger, the methods more complex, and the once-clear boundaries between CV and NLP began to blur.

That transition, she says, was both exhilarating and daunting. “The field was evolving so fast that you had to evolve with it. It wasn’t just about keeping up with the papers—it was about constantly rethinking how you build.”

It was this relentless pace that sharpened her instincts for efficiency, infrastructure, and scalability. She didn’t just want to build models—she wanted to build systems that could run those models faster, cheaper, and more reliably.

Networks, Not Job Boards

In today’s AI job market, one posting can attract a thousand applicants in a single day. “You can’t rely on applying cold,” Kriti says. “Networking is how you stand out.”

It’s not a theory—it’s how she’s built her own career. She speaks candidly about the cultural shift she experienced after moving to the U.S. “In India, companies come to you,” she explains. “Here, you have to hustle. You didn’t grow up in this culture, so you have to learn it. It’s not a complaint—it’s just the reality.”

For her, that reality became a strategy: connecting with peers, staying visible in the field, and building relationships both inside and outside of the workplace. “Networking isn’t just for job hunting,” she adds. “Even if you take time off between college and grad school, or you’re early in your career, those connections matter. They’re what lead to opportunities—especially the early startup roles that never make it to job boards.”

This is why, when asked what advice she’d give to young engineers, she doesn’t hesitate: “Start building your network now. Not later. Now.”

Write Your Way Into the Field

If Kriti has one piece of advice that surprises engineers, it’s this: don’t just build—write.

“Put your work on GitHub,” she says, “but don’t stop there. Write about it. Give your smart take on what you’ve built.”

To her, visibility in AI isn’t just a matter of posting code or linking to a project. It’s about adding context—explaining what you learned, why it matters, and how you approached it. That, she argues, is what turns a side project into a signal of expertise.

She even cites a favorite quote—attributed to Benjamin Franklin: “Either write something worth reading or do something worth writing.” For Kriti, the best engineers do both.

“This is how you get noticed,” she says. “Not just by recruiters, but by the community itself. Writing forces you to think clearly, and it helps other people see how you think. That’s what opens doors.”

The Future of AI (and the Ice Sculpture That Won’t Break)

When Kriti looks ahead, she doesn’t see a future dominated by ever-larger models. She sees a world where AI becomes smaller, faster, and closer to the people who need it most.

“Not every problem needs a 500-billion-parameter model,” she says. “Smaller models that you can fine-tune and run locally—that’s where so much of the opportunity is.” Hospitals are her favorite example: privacy is non-negotiable, so the ability to keep data on-device isn’t just a convenience—it’s a breakthrough.

But what excites her just as much is what most people overlook: the difference between building a demo and building a product.

“Building a cool AI demo over the weekend is like creating an ice sculpture,” she explains. “It’s beautiful, it gets ‘oohs’ and ‘aahs’ in a controlled environment—but it’s fragile and melts under pressure.”

The real challenge? Building something that doesn’t melt. “A real product is like a bridge,” she says. “It has to withstand storms, carry immense traffic, and work flawlessly millions of times without a catastrophic failure.”

This, she adds, is the unglamorous side of engineering: defensive design, not just to make things work—but to make them safe. “You’re preparing so your bridge can get nuked, and you still have to prevent the bad actors from taking down your product or harming users.”

And yet, for all her precision and discipline, Kriti remains deeply optimistic. For her, this isn’t just a field—it’s a craft. A craft that rewards speed and rigor, yes, but also creativity, collaboration, and the quiet discipline of engineers who keep showing up, carving the details, and building something that lasts.

Build Tools That Last

Two days. That’s all it takes to feel behind in AI. But for Kriti Goyal, that pace isn’t a warning—it’s an invitation.

Because while the tools may change every 48 hours, the principles don’t: build things that matter, build them to last, and never stop learning. From fine-tuned hospital models to the “ice sculpture” reliability that production AI demands, her world is proof that speed and craft can coexist—if you’re willing to keep carving.

And maybe that’s the real lesson. AI isn’t just racing forward. It’s also being shaped, one decision, one model, one engineer at a time. And in a field where change never slows down, that kind of steady hand is exactly what endures.

Interested in connecting with machine learning engineer Kriti Goyal ? You can find her here on LinkedIn. Her technical expertise is remarkable—but just as importantly, she’s generous with her insights and a genuinely thoughtful human being.

#ai #machinelearning #mlengineering #techcareers #deeplearning #modeldeployment #trustworthyai #aiforhealthcare #scalingai #privacybydesign #deeplearningwiththewolf #dianawolftorres #kritigoyal