What Are We Actually Designing?
What if the most important output of engineering is no longer just the product, but the knowledge that generates it?
For most of engineering history, designing a product meant defining its physical form.
An engineer translates requirements into geometry, analyses the design, modifies it and eventually tests the physical product. The tools have become enormously more sophisticated, from CAD and simulation to optimisation and additive manufacturing. But the engineer has largely remained the person creating the design.
LEAP 71, a Dubai-based engineering startup, is exploring a different possibility.
When Geometry Becomes an Output
LEAP 71 calls its approach computational engineering.
Its Noyron system encodes engineering knowledge, physics and manufacturing constraints computationally. Given a set of requirements, it can generate the geometry of a complex physical machine.
Traditionally, geometry is one of the principal ways engineers express a design. With computational engineering, more of the decisions behind that geometry can be expressed in the system that generates it.
The geometry becomes an output.
LEAP 71 has demonstrated the approach with rocket engines that were generated computationally, additively manufactured and hot-fire tested.
The physical test remains essential. But its results can also inform the computational model that generates future engines. After a 2024 hot-fire test, for example, LEAP 71 found that the pressure drop in the cooling channels was higher than predicted and updated Noyron’s cooling-channel logic accordingly.
The loop looks different:
engineering knowledge → computational model → physical product → test → improved computational model
A Body of Engineering Knowledge
Engineering organisations contain enormous amounts of knowledge. It is distributed across people, documents, standards, equations, software and previous designs. Much of it has to be interpreted and applied again each time a new product is developed.
Computational engineering offers another way for that knowledge to accumulate.
If engineering knowledge is encoded in a system that generates designs, it can be applied repeatedly. When testing reveals something new, that knowledge can feed back into the system and influence future designs.
There are also early signs that computation can contribute to the other side of this loop: producing new engineering knowledge. In a 2026 Nature Communications study, researchers used interpretable machine learning to predict degradation in NiTi shape-memory alloys and to identify physical mechanisms associated with that degradation under different loading conditions. The computational model therefore became a tool for learning about the material as well as predicting its behaviour.
This suggests a broader feedback loop. Engineering knowledge can be encoded into computational systems, while those systems can help extract new knowledge from experiments and data. That knowledge can then become part of the models used for future engineering work.
The result is a reusable and potentially improving body of engineering knowledge.
What Are We Actually Designing?
LEAP 71’s rocket engines therefore point to something broader than automated geometry.
The engine is one output. The computational engineering knowledge capable of generating it is another.
For most of engineering history, we have designed products and then tried to capture what we learned from designing and testing them.
Computational engineering offers the possibility of closing that loop: what we learn from one product becomes part of the knowledge used to create the next.
References
- LEAP 71, Noyron — a foundational Large Computational Engineering Model
- LEAP 71, LEAP 71 hot-fires 3D-printed liquid-fuel rocket engine designed through Noyron Computational Model (2024)
- Hu et al., An interpretable learning framework for exploring superelastic degradation of NiTi shape memory alloys using multimodal data, Nature Communications (2026)
AI helped to research and write this article.