Alternative image

When the lab drives itself: The story behind Evonik’s award-winning platform

Evonik's Self Driving Labs won the 2026 CPHI Pharma Award in the "Digital Transformation" category for its application in the development of active pharmaceutical ingredients (APIs). The platform uses artificial intelligence to design, run and evaluate experiments for API process development. It also builds on years of work by teams across Evonik, from 3D-printed reactors to data science.

Read: ca. 4 min

by Lucas Cusin & Thilo Krapfl


What if a lab could keep experimenting around the clock? What if challenging chemical processes could be carried out more simply in a reactor tailored to the chemistry? What if pharmaceutical processes could be developed faster – and be more efficient from the start? What if every chemist and technician could use AI-driven tools through an easy-to-use interface? Evonik’s Self Driving Labs is designed to make all this possible.

Self Driving Labs is a process development service for pharmaceutical customers worldwide. It combines artificial intelligence with automated lab equipment to plan, run and evaluate experiments with minimal manual intervention. For customers, this means faster development, better data for scale-up decisions, and processes that can use less material and generate less waste.

For applying Self Driving Labs to active pharmaceutical ingredient (API) development, Evonik’s Health Care business won the 2026 CPHI Pharma Award in the Digital Transformation category. But this success did not happen overnight. It builds on capabilities that several teams across Evonik developed over many years: online measurement and process analytical technology, simulation, 3D-printed reactor design, machine learning and an easy-to-use software interface. This story shows how these pieces came together.

The pharmaceutical development process: defined processes, hard to change

In early API development, the priority is to deliver small quantities quickly. As a molecule moves through clinical phases, the focus shifts to establishing a commercially viable manufacturing process. Once that process is defined, however, it becomes harder to change without risking product quality, adding cost or causing delays.

Conventional process development can be slow and resource intensive. Experiments oftentimes are planned and run one after another, by hand. Developing a single reaction and workup can take several weeks of expert work, produce only limited analytical data and depend heavily on expert judgement. At the same time, market trends show that pipelines are expanding beyond small molecules to complex modalities such as peptides and oligonucleotides, and conventional methods struggle to keep up.

Self Driving Labs addresses this by digitalizing the core development workflow, from experiment design and execution to data capture, analysis and decision-making.

A new way of developing processes for pharma

For pharmaceutical companies, the value of Self Driving Labs lies in reaching a robust, scalable process earlier, ideally in the first generation of a manufacturing route. Richer data from the start supports faster, lower-risk scale-up decisions and makes costly late-stage process changes less likely.

"Traditional development often relies on batch reactors and established approaches," says Michael Nonnenmacher, director of global business development for differentiating technologies for Drug Substance at Evonik Health Care. "With Self Driving Labs, we introduced a fundamentally different development methodology. We optimize both the reaction conditions and the reactor design, enabled by machine learning and advanced manufacturing."

The results are measurable. In one industrial use case, a Grignard reaction run in continuous flow, experimental throughput rose from around two manually controlled experiments to as many as 100 autonomous experiments per day. In a chromatographic separation developed with the company Agilent, a task that normally takes a week of expert-led experimentation was completed autonomously overnight, with equal or better performance.

Where it started: the 3D-PROCESS project

Many of the foundations of Self Driving Labs were laid in the project 3D-PROCESS. This publicly funded German consortium project was supported by the Federal Ministry for Economic Affairs and Climate Action (BMWK). In the project, Evonik developed reactor systems that are digitally designed, 3D-printed and optimized using process data, simulation and artificial intelligence.

On Evonik's side, the work was led by the Simulation and Additive Manufacturing competence center, SAM 3D, headed by Senada Schaack. 3D-PROCESS combined additive manufacturing, simulation (including fluid dynamics and mechanical simulation), machine learning, sensing and automated process control. "We are completely rethinking the processes in chemical engineering toward precisely shaping molecules to reduce side products, solvent use, and enhance the sustainability of processes," Schaack said of the project.

The Health Care workstream was one of several in the wider initiative. Its original concept was a "self-learning reactor" that could optimize process conditions on its own through machine learning. "The project showed that we could accelerate experimentation by up to a factor of up to 100, cutting development time while making processes more efficient," says Schaack. What Evonik learned in 3D-PROCESS shaped its internal roadmap and enabled it to develop Self Driving Labs independently for its pharmaceutical use cases.

Reactor design: small, tailored and 3D-printed

As part of 3D-PROCESS, the teams created 3D-printed flow reactors from stainless steel powder. Each reactor's internal geometry, such as very thin channels for the reaction mixture or cooling medium, is designed for the specific chemistry. This makes it possible to control highly exothermic reactions precisely, reduce side-product formation and use less solvent.

Miniaturized, structured reactors like these support the wider shift from batch to continuous pharmaceutical processing. "The compactness of these reactor systems is a real advantage," says Hendrik Rehage, the Project Lead of 3D-PROCESS at Evonik. "A small setup can achieve significant production capacity without requiring the space of a conventional plant."

In Self Driving Labs, this capability became part of the optimization loop. In the Grignard use case, the team modified the reactor geometry step by step using 3D printing.

Bayesian optimization: the algorithm that decides what to test next

At the heart of Self Driving Labs is a machine-learning method called Bayesian optimization, run through BoFire, an open-source optimization framework. Johannes Dürholt, main contributor to BoFire from Evonik’s R&D digitalization unit, explains: “The algorithm learns from the results of every experiment and proposes new experimental conditions. This way, each experiment guides the next toward defined goals, such as higher yield, less waste, or a combination of these goals.” 

In closed-loop mode, process data are collected continuously and fed straight back to the algorithm, so the system can run around the clock with very little human supervision. The algorithm operates only within a predefined, safety-assessed operating window, and the system includes automated interlocks. The approach also tries options that people might rule out. In the Grignard case, the algorithm proposed operating at the upper limit of the allowable temperature, which ultimately led to better performance. "During optimization, the algorithm identified reaction conditions outside the range we would typically have tested – and they turned out to be the best," says Benjamin Häfner, director of process technology at Evonik Health Care.

Bayesian optimization can also be used in an open-loop setup. Here, laboratory staff run the experiments and enter the data, while the algorithm guides development more efficiently.

AIChemBuddy: advanced optimization for every lab

Powerful algorithms only make a difference if chemists and technicians can use them without specialist data science expertise. That is the role of AIChemBuddy, Evonik's in-house R&D data science platform and the user interface of Self Driving Labs. It was developed by a team coordinated by Thomas Asche from Evonik's R&D digitalization unit. It helps researchers plan and optimize experiments when formulations, reactions or processes involve many interacting variables.

With AIChemBuddy, users can explore and visualize experimental data, build models, run virtual experiments and get proposals for the most promising next experiments. "We propose experiments on the basis of the desired outputs," says Asche. "This optimization is the core element of the platform." Researchers keep scientific judgement and the final decisions, and full digital traceability gives them transparency and control. The digital traceability of every experiment is especially important for the pharmaceutical industry, where it can be a real differentiator for regulatory filings and tech transfer.

The platform has already proven its value. A team in Evonik's Health Care R&D was developing a production process for a new lipid used in lipid nanoparticles. AIChemBuddy helped them narrow down the reagents, optimize the synthesis and find a way to significantly increase the product purity within a few weeks.

Flexible building blocks for the future

Self Driving Labs is not tied to one site or reactor technology and is being extended across Evonik's global Health Care organization. It is used both in new development projects and to optimize existing products within the regulatory framework.

Fully closed-loop operation is not the only option. Customers can choose the components that suit the size and needs of their project. "Fully autonomous optimization is not always the fastest or most practical solution, especially for smaller projects where the setup effort may outweigh the benefits," says Häfner. "We apply selected elements flexibly to meet specific customer requirements – and still accelerate development." 

Work on the reactors continues too. Some goals of the original project, such as using algorithms to optimize reactor geometry, were not yet within reach at the time and have since moved forward. "Today, we can iteratively redesign and test reactor configurations – something that wasn't possible a few years ago," says Rehage.

The CPHI award shows what happens when expertise in chemistry, engineering, automation, IT and data science comes together. For pharmaceutical customers, the result is a faster path to robust, scalable and efficient API processes, starting with the very first generation.


ANY QUESTIONS? JUST CONTACT US.

"Building a more sustainable future for oral drug delivery requires the right balance of performance, regulatory readiness, and environmental responsibility. Our global team of oral drug delivery experts is ready to support you with innovative excipients, sustainability insights, and formulation expertise to help bring your next generation of drug products to life."

Felix Hofmann
Director Formulation & Application Services

Send me a message
Alternative image

More about our offerings for oral drug delivery

Alternative image

EUDRAGIT® functional polymers

Alternative image

EUDRACAP® functional capsules

Alternative image

Formulation & development services