How a Korean biotech startup is using AI to move drug discovery from trial-and-error to precision design
A close up of a protein structure model. PHOTO: UNSPLASH
For decades, drug discovery has relied on trial and error, with scientists testing thousands of molecules to find one that works. Galux, a South Korean biotech startup, is changing that by using AI to design proteins from scratch. This method, called “de novo” design, makes it possible to build precise new therapies instead of searching through existing ones.
The company recently announced a US$29 million Series B funding round, bringing its total capital to US$47 million.This significant investment attracted a substantial roster of institutional backers, including the Korea Development Bank (KDB), Yuanta Investment, SL Investment and NCORE Ventures. These firms joined existing investors such as InterVest, DAYLI Partners and PATHWAY Investment, as well as new participants including SneakPeek Investments, Korea Investment & Securities and Mirae Asset Securities.
At the core of the company’s work is a platform called GaluxDesign. Unlike many AI tools that only predict how existing proteins fold, this system uses deep learning and physics to create entirely new therapeutic antibodies. This “from scratch” approach lets the team go after so-called “undruggable” proteins. These are targets that traditional small-molecule drugs can’t reach because they lack clear binding pockets. By designing proteins to fit these complex shapes, Galux aims to unlock treatments that have stayed out of reach for decades. And that’s exactly why investors are paying attention.
The pharmaceutical industry is actively looking for faster and more efficient ways to develop new drugs, and Galux is built for exactly that. The company connects its AI platform directly to its own wet lab, where designs can be tested in real time. Each result feeds straight back into the system, sharpening the next round of models. This continuous loop speeds up discovery and improves precision at every step. It’s also why partners like Celltrion, LG Chem and Boehringer Ingelheim are already working with Galux.
Galux is no longer just trying to make drugs that stick to a target. The company now wants its AI to design medicines that actually work in the body and can be made at scale. In simple terms, a drug has to do more than bind to a disease—it must be stable, safe and strong enough to change how the illness behaves. Galux is moving into tougher targets such as ion channels and GPCRs. These play key roles in heart function and sensory signals. Ultimately, the goal is to show that AI-driven design can turn complex biology into real treatments. And instead of hunting blindly for a solution, the team is building exactly what they need.
Keep Reading
A look at how motivation, not metrics, is becoming the real frontier in fitness tech
Updated
February 7, 2026 2:18 PM
A group of people running together. PHOTO: FREEPIK
Most running apps focus on measurement. Distance, pace, heart rate, badges. They record activity well, but struggle to help users maintain consistency over time. As a result, many people track diligently at first, then gradually disengage.
That drop-off has pushed developers to rethink what fitness technology is actually for. Instead of just documenting activity, some platforms are now trying to influence behaviour itself. Paceful, an AI-powered running platform developed by SportsTech startup xCREW, is part of that shift — not by adding more metrics, but by focusing on how people stay consistent. The platform is built on a simple behavioural insight: most people don’t stop exercising because they don’t care about health. They stop because routines are fragile. Miss a few days and the habit collapses. Technology that focuses only on performance metrics doesn’t solve that. Systems that reinforce consistency, belonging and feedback loops might.
Instead of treating running as a solo, data-driven task, Paceful is built around two ideas: behavioural incentives and social alignment. The system turns real-world running activity into tangible rewards and it uses AI to connect runners to people, clubs and challenges that fit how and where they actually run.
At the technical level, Paceful connects with existing fitness ecosystems. Users can import workout data from platforms like Apple Health and Strava rather than starting from scratch. Once inside the system, AI models analyse pace, frequency, location and participation patterns. That data is used to recommend running partners, clubs and group challenges that match each runner’s habits and context.
What makes this approach different is not the tracking itself, but what the platform does with the data it collects. Running distance and consistency become inputs for a reward system that offers physical-world incentives, such as gear, race entries or gift cards. The idea is to link effort to something concrete, rather than abstract. The company also built the system around community logic rather than individual competition. Even solo runners are placed into challenge formats designed to simulate the motivation of a group. In practice, that means users feel part of a shared structure even when running alone.
During a six-month beta phase in the US, xCREW tested Paceful with more than 4,000 running clubs and around 50,000 runners. According to the company, users increased their running frequency significantly and weekly retention remained unusually high for a fitness platform. One beta tester summed it up this way: “Strava just logs records, but Paceful rewards you for every run, which is a completely different motivation”.
The company has raised seed funding and plans to expand the platform beyond running, walking, trekking, cycling and swimming. Instead of asking how accurately technology can measure the body, platforms like Paceful are asking a different question: how technology might influence everyday behaviour. Not by adding more data, but by shaping the conditions around effort, feedback and social connection.
As AI becomes more common in consumer products, its real impact may depend less on how advanced the models are and more on what they are applied to. In this case, the focus isn’t speed or performance — it’s consistency. And whether systems like this can meaningfully support it over time.