They needed parts that were lighter, stronger and more efficient, but development was slow, costly and creatively stuck. Engineers drew in classic CAD, built expensive prototypes and tested them in wind tunnels, so each iteration took months and mostly produced small tweaks to what already existed.
Every development cycle ran the same slow loop: model a part in CAD, build a costly physical prototype, test it in a wind tunnel or stress lab, and start again. Each turn took months, and the result was usually a marginal improvement on an existing design.
The bigger limit was human. Engineers, however brilliant, are anchored to the shapes they already know, so genuinely new geometries, the ones a person would never think to draw, stayed undiscovered.
And it was expensive. Producing many physical prototypes and running lab tests that failed, again and again, before reaching an optimal result, made every part slow and costly to get right.
Give the AI the hard constraints, let it generate and simulate thousands of options, and hand the winners to an engineer.
We stood up a generative-design platform that takes the engineers' hard constraints, maximum weight, thermal loads, permitted materials and the manufacturing methods available, as the brief the design has to satisfy.
The algorithm autonomously produces thousands of distinct design variations that meet the requirements, including radical geometries no engineer would have drawn by hand.
Each variation is put through complex physics simulations in a virtual environment, so the testing that used to mean a physical prototype now happens in software, in a fraction of the time.
The best-performing models are surfaced to the engineers for a final human polish and judgement call, so the AI widens the search and the person still owns the design that ships.
The chosen design goes directly to industrial 3D printing, so the optimised geometry is manufacturable as designed rather than compromised back into something a traditional process can cut.
The R&D team moved from technical drawing to defining the problem and steering the algorithms, which is where their expertise creates the most value.
The search got wider and faster, and the engineer still signed the design.
Component weight fell by as much as 30% while structural strength held, which translated into large fuel and energy savings for the company's own customers downstream.
Replacing physical trials with fast virtual simulation cut engineering development from months to a few weeks, and shifted the R&D team from drawing parts to setting the strategy that guides the search.
Traditional design is bounded by what an engineer can picture and how fast a prototype can be built and broken. The value here is giving the AI the real physical constraints so it can explore a space no person could, proving each option in simulation, and leaving the final call to an engineer. It is the same constraints-first, human-validated discipline behind our bid-pricing work, aimed at the design of the part rather than the price of the tender.
A contractor kept winning loss-making tenders. Grounding every bid in real field costs with a hard margin floor added about 3.5% gross margin and cut pricing from three weeks to 48 hours.
Read the case → Computer vision · AgricultureApproving farm credit meant sending a surveyor out for weeks. Computer vision on satellite imagery scored crop health and collateral, taking the company from dozens of applications a week to 25,000 farms a day.
Read the case →
Reef TRH