No open roles right now

Drug discovery is one of the most important engineering problems in the world, yet the way it’s done is still painfully slow. A promising molecule can spend years moving through cycles of prediction, synthesis, testing, and redesign, and most candidates fail somewhere along that path.

Many of those failures leave signals long before the experiment that confirms them. A molecule may have poor absorption, unacceptable toxicity, difficult synthesis, or an unstable binding mode. The challenge isn’t that these questions are impossible to answer. It’s that answering them requires too much time, too many tools, and too many handoffs. We think that should change.

KemJet is building AI for drug discovery, from the models themselves to the product that puts them into scientists’ hands. We build foundation models for molecular reasoning, prediction, and design, then turn them into workflows that help researchers move from an idea to a decision in minutes instead of weeks.

Our goal isn’t to replace scientists. Our goal is to remove the friction between asking a question and getting an answer worth trusting.

A fast prediction is useless if nobody knows when it’s wrong. Every model has blind spots. Every prediction has uncertainty. We think the next generation of scientific AI will be judged less by how confidently it speaks and more by how honestly it admits what it doesn’t know.

That’s why we invest heavily in calibration, confidence estimation, domain awareness, and evidence. When KemJet makes a claim, it should also communicate how much confidence deserves to be placed in it. Trust isn’t a feature. It’s part of the model.

Powerful models deserve better interfaces. Scientists shouldn’t need to become machine learning engineers to explore an idea. They shouldn’t spend half their day moving files between disconnected software or waiting for someone else to run an analysis.

We’re building an experience that feels immediate: generate molecules, predict ADMET properties, evaluate liabilities, explore binding hypotheses, plan synthesis, and iterate without breaking your train of thought. The interface isn’t decoration. It’s part of the research loop.

We want to shorten the distance between curiosity and evidence. The best version of KemJet is one where a researcher has an idea in the morning, tests it before lunch, learns something real, and moves to the next iteration before the day is over.

Drug discovery will always be hard. We’re building the intelligence, the infrastructure, and the experience that make it faster to find what actually works.

Write to thekemjet@kemjet.org with what you have built and what you would want to work on. Something you have made tells us more than a CV does, though send both if you like.

We read everything and reply to anything we can act on. We will not keep you in a pipeline that does not exist: if there is nothing, we will say there is nothing.

We hold what you send for twelve months so we can come back to you when something opens, then delete it. Ask us at admin@kemjet.org to remove it sooner and we will. The privacy policy covers the rest.