Data Science Intern — Fermentation Modelling & Soft Sensors

We’re building the protein breweries of the future. Our process is up to 5,000× more land‑efficient than animal or plant protein. In the coming years, we’ll run our first commercial facilities producing thousands of tonnes of protein.

This internship is about understanding what happens inside them. A fermentation is a living system you can barely see into, and almost everything we decide about it rests on inferring the rest.

Our fermentation is a bacterial culture feeding on a renewable liquid feedstock, in a stainless steel vessel, for days at a time. The things you actually want to know — how much biomass is in there, how much product, whether the culture is healthy or about to stall — are the things you cannot measure while it runs. What you get continuously is the periphery: off‑gas, feed rate, dissolved oxygen, pH, temperature, vessel weight, heat.

The variables that matter arrive hours later from a lab assay, if at all.

So you infer. Everything downstream of that — control, scale‑up, deciding what to run next — depends on how well.

Our dataset is dozens of fermentations, not millions of rows, and each one costs weeks of plant time to produce. The work that follows from that is mechanistic modelling, Bayesian inference, statistics that takes uncertainty seriously, and design of experiments that earns its keep because the experiments are expensive. You reach for the process before you reach for the algorithm.

The other half of the job is being understood. You’ll be explaining what your model found to the process engineers and fermentation scientists who have to act on it, in terms of the process rather than the method. They’re in the same building, they know things your data doesn’t, and they will tell you when you’re wrong.

One central project, yours end to end. See three ideas below: which one you take depends on where you’re strongest and what the plant needs when you arrive, and we’ll work that out with you.

Build a soft sensor.Infer the unmeasurable from the measurable: biomass and product concentration, live, from off‑gas, feed rate and the rest of the online signals. State estimation on a real process. The end state is the interesting part — a good estimate doesn’t stay in a notebook, it goes into Brew Control and closes the loop on a vessel in this building.

Model the organism.A kinetic model of our strain, fitted against our run history, that predicts what a fermentation will do before it runs. Mechanistic where the biology is understood, data‑driven where it isn’t, and honest about which is which. The deliverable is a model the fermentation team trusts enough to plan against.

Design the next experiment.When a single run costs a fortnight, choosing the right one is worth more than any analysis of the last one. Design of experiments over the process parameters, with a model of the process as the surrogate, choosing sequentially as results come in. The deliverable is the run we do next.

The bar is the same whichever you take: someone bets a fermentation on it — a real run, planned differently because of what your model said.

You model from first principles.The instinct we’re screening for is reaching for the process before the algorithm.

Statistics, properly.Uncertainty, priors, mixed effects, experimental design. You should be able to say what your model doesn’t know.

Python, scientifically.The numerical and statistical stack, not just the ML one.

You’ve met real data.Messy, small, expensive, gappy, with a sensor that was miscalibrated for three weeks in the middle.

You can talk to a process engineer.Explaining a result in terms of the process, to someone who knows the process better than you do, is half of this job.

Fast learner.Excited to dive into fermentation biology and bioprocess engineering.

Mission driven.Rewriting how the world makes protein genuinely excites you.

Bonus points for: bioprocess or chemical engineering, state estimation and control theory (Kalman filters and friends), Bayesian methods, time‑series at scale, a biology, physics or systems biology background, enough software engineering to ship what you build, previous…