This episode of Bright Founders Talk by Temy features an insightful conversation with Handol Kim, CEO and co-founder of Variational AI, a company redefining drug discovery through generative AI. Based in Vancouver, Variational AI is pioneering a platform that designs small-molecule drug candidates for biotech and pharmaceutical partners.

In this interview, Handol shares his unconventional journey from a 25-year career in software, machine learning, and quantum-adjacent technologies into the complex world of biotechnology. He reflects on how his team transitioned from a quantum computing company to building a startup focused on solving one of the hardest challenges in science. The discussion uncovers how generative AI can dramatically improve the efficiency and success rate of early-stage drug discovery.

Handol also offers a candid look at the realities of entrepreneurship, including failures, lessons learned, and the resilience required to build a company from scratch. Drawing parallels between tech and biotech, he highlights both the opportunities and the communication gaps between disciplines. This conversation provides a compelling perspective on innovation at the intersection of AI and life sciences, and what it takes to push boundaries in both fields.

From Quantum to Chemistry: How Handol Kim Is Rewriting Drug Discovery with AI

Handol’s story doesn’t start in a lab — it starts deep in tech. With over two decades in software, machine learning, and even quantum-adjacent work, biotech wasn’t exactly the obvious next step. But sometimes the best ideas come from being pushed out of your comfort zone. When his team’s unit at a quantum computing company was shut down, they were told to go build something — just not quantum. So they did exactly that. What followed was the birth of Variational AI, a company tackling one of the toughest problems out there: discovering new drugs using generative AI.

“I’m an immigrant to biotech,” Handol says — and that outsider perspective turned out to be a strength. Coming from tech, he and his team approached drug discovery differently, applying cutting-edge machine learning to a field that traditionally moves slow and cautiously. But it wasn’t all smooth sailing. Switching industries meant relearning the language — sometimes literally. Same words, different meanings, and that small gap? That’s where confusion (and mistakes) happen. Still, those early missteps became part of the journey, shaping how the company evolved.

And then there’s the reality of drug discovery itself — a process that makes startups look easy. We’re talking 10–12 years, billions of dollars, and only about a 10% chance of success once human trials begin. As Handol puts it, “It’s the most expensive lottery ticket in the world.” That’s exactly the problem Variational AI is trying to solve — focusing on the earliest stage, where better decisions can save years of work and massive costs. Instead of guessing which molecules might work, their platform explores an almost unimaginable chemical space, helping scientists find better candidates faster — and maybe, just maybe, tilt the odds a little more in our favor.

Exploring the Unknown: How AI Is Unlocking the Dark Matter of Drug Discovery

Imagine trying to find a needle in a haystack — except the haystack has 10⁶⁰ possibilities. That’s the reality of drug discovery. Chemists aren’t just creating molecules out of thin air; they’re navigating an almost infinite “chemical space” where most options are useless — or worse, harmful. So instead of starting from scratch, they play it safe, tweaking existing drugs step by step in a slow, expensive loop of design, testing, and analysis. It works, but it’s painfully incremental and limits how far you can actually go.

That’s where Handol and his team flip the script. Instead of staying in one small corner of chemical space, Variational AI is built to explore it — properly. Using generative AI, they create entirely new molecules, much like how image models generate visuals that have never existed before. The idea is simple but powerful: better exploration leads to better starting points, and that means less time and money spent refining them later. As Handol puts it, “If you explore better, you exploit less.” And in an industry where years and billions are on the line, that shift can change everything.

But here’s the catch — no matter how smart the model is, nothing matters until it works in the real world. AI can suggest promising molecules, but they still have to be physically created and tested in the lab. That’s the ultimate truth in biotech. So while Variational AI pushes the boundaries of what’s possible digitally, it stays grounded in reality: experimentation is king. The magic happens when cutting-edge AI meets hard science — and together, they start turning the impossible into something just within reach.

“Train on Garbage, Get Garbage”: The Hard Truth About AI in Drug Discovery

For a moment, it looked like they had cracked it. Early on, Handol’s team was hitting state-of-the-art results, outperforming others and presenting at top-tier conferences like NeurIPS. It felt like a breakthrough — maybe even the breakthrough. But reality hit fast. Models that looked perfect on paper didn’t always hold up in the lab. The reason? The data. As Handol bluntly puts it, “Train on crap, get crap.” In a field where data is scarce, noisy, and expensive to generate, even the smartest AI can only go so far.

That’s why, despite all the hype around AI, Handol keeps things grounded. No prediction matters until it survives real-world testing. You can have beautiful models, strong benchmarks, and impressive simulations — and still fail the moment you synthesize the molecule. It’s a humbling process, and it forces a shift in mindset: AI isn’t magic, it’s a tool. A powerful one, yes, but still dependent on the messy, imperfect reality of biology and chemistry. And in that reality, surprises are the norm, not the exception.

What makes things even more complex is that Variational AI doesn’t chase specific diseases — it chases targets. Biology is just too unpredictable. Instead of trying to solve cancer or eczema directly, they focus on the underlying proteins, acting more like precision shooters than broad strategists. “We’re like snipers,” Handol explains — highly accurate, but only if you point them in the right direction. And that’s where collaboration comes in. Biotech partners bring the biological insight, Variational brings the AI firepower. Together, they’re navigating one of the most complex frontiers in science — all while keeping an eye on an even bigger force: the relentless, and slightly terrifying, pace of AI itself.

AI Hype vs Reality: Why Drug Discovery Still Plays by Different Rules

The AI boom is impossible to ignore — and Handol feels it just like everyone else. The pace of innovation, especially with large language models, is not just exciting, it’s borderline overwhelming. Entire industries are starting to question their future, from law to accounting. But history tells us something important: people adapt. Just like designers didn’t disappear with Photoshop, most professions will evolve rather than vanish. Still, in drug discovery, things don’t move at “AI speed” — and for a good reason.

Unlike other AI domains, you can’t just throw more data and GPUs at the problem and expect magic. In chemistry, the data simply isn’t there. It’s limited, expensive, and painfully slow to generate. While language models train on trillions of tokens, drug discovery models work with datasets that are tiny by comparison — and growing only incrementally. That changes the game completely. As Handol puts it, “You can’t just make the model bigger.” It forces teams to think smarter, not just scale harder, and to be incredibly selective about what data they use.

And then there’s the bigger reality check: even with all the AI progress, biology remains brutally complex. Solving “all diseases” sounds great in headlines, but the truth is far messier. A drug might work perfectly — until it doesn’t, or worse, harms a subset of patients. That’s why Handol stays cautiously optimistic. AI will absolutely move the field forward, but it won’t replace the hard parts overnight. In the end, it’s still about improving patient outcomes, balancing innovation with safety, and navigating a system where science, economics, and regulation all collide.