Machine Learning Data Scientist

Role summary:

At Differential Bio, our mission is to unlock the potential of the bioeconomy by fixing how biological R&D and scale-up are done.

We’re looking for a Machine Learning Data Scientist to join our growing cutting-edge modelling team. You will develop novel machine learning algorithms and mechanistic models of bioprocess systems and push them from research proof-of-concept into production-ready commercial use.

This role is focused on research, including creating hybrid models that combine mechanistic and machine learning approaches to predict how microbial fermentations behave, and that keep working when we move to a new organism, a new sensor, or a larger reactor. You’ll play an instrumental role on two existing grants and work with leading academic groups and industrial partners on pilot- and production-scale data. In addition, you’ll develop new research projects and drive them to completion and application on customer projects.

We’re interested in researchers who want their work to end up in real bioreactors and are excited about working at an early-stage startup. You should be comfortable owning hard, open-ended problems where nobody has the answer yet, making evidence-based decisions under uncertainty, and shipping real solutions.

How you’ll work

• You’ll collaborate with technical leads and academic/industry external partners to design and execute research that meets our commitments in both consortia and strengthens our own platform.

• You’ll own your research projects from scoping to delivery, with support from our modelling lead but with autonomy and independence.

• You’ll work closely with our wet lab, dry lab and software teams, and identify which data you need to generate from our robotic lab.

• You can work remotely, provided you’re based within ±2 hours of Central European Time (CET), with occasional travel

Key responsibilities:

Hybrid bioprocess modelling

• Develop hybrid mechanistic/ML cell response models that predict fermentation outcomes (growth, titer, yield, productivity) from process and sensor data.

• Fuse time-series signals from process analytical technology (Raman, NIR, dielectric spectroscopy, dissolved oxygen, off-gas) with offline analytics.

• Build mechanistic backbones (e.g., kinetic/ODE models, genome-scale metabolic models) that constrain learning and generate synthetic training data.

• Develop machine learning surrogate models for computationally costly mechanistic simulations.

• Quantify uncertainty so that models can be trusted in high-stakes production settings.

Transfer across organisms and scales

• Design models that carry over from benchtop to pilot and production scale, and across various microbes including bacteria and yeast.

• Use transfer learning and representation learning to make good predictions for new strains and processes with little data.

• Develop scale-transferable reactor characterisations and predictive KPI models for scale-up.

Real-time Model-Predictive Control

• Build the AI/ML layer for Nonlinear Model Predictive Control (NMPC) and reinforcement-learning-based feeding strategies to optimize titer, rate, and yield dynamically.

• Fuse real-time data from Process Analytical Technology (PAT) and soft sensors (e.g., Raman, NIR) into dynamic state estimators to drive closed-loop bioreactor actuators (feed, air, agitation).

• Prototype simpler, deployable steps along the way, such as deviation detection and operator alerts

• Design control frameworks to proactively resolve metabolic bottlenecks and robustly account for model uncertainty, batch-to-batch variability, and inherent bioprocess stochasticity.

Qualifications:

• PhD in machine learning, bioengineering, computational biology, or a related field, or a Master’s degree with 3+ years of relevant industry experience.

• Strong research independence; ability to scope and execute novel scientific work from start to finish.

• Demonstrated evidence of research contributions, including (but not limited to): peer-reviewed journal or conference papers, conference attendance, open-source contributions.

• Strong Python and modern ML tooling (e.g., PyTorch/JAX, scikit-learn).

• Hands-on experience building models of dynamical systems or time-series data, ideally combining mechanistic and data-driven approaches.

• Experience working with frontier coding models, including multi-agent orchestration.

• Clear written and verbal communication, including with non-ML partners.

Nice-to-have:

• Experience with bioprocess or fermentation data, ideally from bioreactors at more than one scale

• Experience with Raman, NIR, or other spectroscopic data and chemometrics (e.g., PLS)

• Background in model-predictive control, reinforcement learning, or optimal control

• Experience with transfer learning, domain adaptation, or representation learning on biological data

• Familiarity with genome-scale metabolic models or kinetic modelling

• Experience with Bayesian methods, uncertainty quantification, or design of experiments

• Experience in collaborative or publicly funded research projects (e.g. Horizon Europe, Eurostars)

Benefits:

• Competitive compensation through salary + stock options; we are looking to build a small but mighty team and think about compensation along those lines;

• Build your research profile: Work on open problems with leading European academic and industrial partners, with the intention to present and publish your work;

• A beautiful office in the heart of Munich, or remote work option with occasional travel

• eGym WellPass membership