Research / Open questions
Grounded in research.
Built for industry.
We’re investigating how industrial intelligence could respond to unfamiliar faults, adapt as conditions change and learn across batches. Computer science, machine learning and process control provide the foundations for this work.
Our approach to research
Physical process models give a way to describe how a process behaves. Machine learning offers ways to learn from observations and experience. AI reasoning can help interpret a situation in context. We’re exploring their contribution to a purpose-built industrial system, with LLMs as one component. The ambition extends beyond a standalone or fine-tuned language model; it depends on understanding and acting within a physical process.
Questions guiding our work
Responding to unfamiliar faults
A useful system needs to handle situations that were not anticipated during development. We’re asking how it could recognise that its current understanding is incomplete and still make a useful contribution to recovery. This is a research objective, not a capability established by the work so far.
Adapting to changing conditions
A process can change between batches as well as within one. We’re investigating how intelligence could remain useful as those conditions evolve, reducing the need to rebuild a solution for each variation. The challenge includes knowing when past experience is relevant and when it needs to be reconsidered.
Learning usefully across batches
Recording an incident is different from learning something that improves a later response. We’re exploring what useful learning would mean for a manufacturing process: experience that can inform future decisions while remaining sensitive to the circumstances of the current batch. This remains part of the work ahead.
A place to begin
A first environment for research.
Our starting point is a synthetic bioreactor, where oxygen availability, nutrient supply and changing process conditions provide a concrete setting for development. It gives us room to investigate interacting behaviour before an industry trial. Simulation can support the development of ideas and software; industrial validation requires separate evidence from real processes, their equipment and the people responsible for them.
Research grows through dialogue.
We welcome researchers and manufacturing specialists who can sharpen the questions, challenge assumptions and help connect this work with a useful industrial problem.