Michelle
Lee

AlphaFold cracked folding. She asked who runs the experiments.

In 2021, AlphaFold 2 landed and the world celebrated a solved problem. Lee, then heading toward a quiet academic career, saw the unsolved one. If AI could predict how a protein folds, why not what it does - its function, its behavior, its role in a disease? The answer was uncomfortable: there was not enough data. Not because nobody was collecting it, but because collecting it means going physically into the lab and running experiments that are, in her words, "incredibly laborious, repetitive."

Labs can only run as many experiments as they have scientists. Scientists have working hours. And when a scientist leaves, the craft leaves with them - the exact angle and depth of a pipette tip, the timing, the feel. Lee reframed the entire bottleneck in one phrase that has since become her thesis: data is a robotics problem.

So she did the thing that makes founders founders. She left a faculty post at NYU - assistant professor of computer science and electrical and computer engineering - and started building the infrastructure to fix it. The Stanford talk she later gave about the decision has a fitting title: "From Conviction to Company."

It records the angle of the pipette

Here is the detail that tells you Lee thinks differently. Most automation tries to remove the human from the experiment. Medra tries to capture what the human knows and never wrote down. The system logs the exact angle and depth of the pipette tip, the precise timing, the small physical decisions that no protocol ever specifies. Because, as Lee puts it, "the artisanal nature of science is actually what makes certain experiments work and others fail."

Biology is variable and stubborn. The same protocol fails in one pair of hands and sings in another. Medra's bet is that the difference is not magic - it is data nobody bothered to capture. So the robots capture all of it, every run, and turn craft into something a model can learn.