The Uretyco AI co-pilot is built to help engineers and procurement teams reach a good quote faster. It is not a magic black box. This article describes exactly what it does, where its limits are, and how we handle the data it sees.
What the co-pilot does
- Reads the topology output of your uploaded model and answers questions about it.
- Suggests appropriate processes and materials based on intended use.
- Flags DFM risks (thin walls, deep pockets, undercut threads) before the quote is final.
- Estimates impact of changes (relax this tolerance, swap to a different alloy).
- Produces a rationale for the suggested quote, in plain English.
What it deliberately does not do
- It never modifies your CAD model.
- It never finalizes an order on your behalf.
- It never trains on your CAD geometry without explicit, granular consent.
- It never ranks suppliers without showing the reasoning.
Data boundaries
The co-pilot sees only the metrics produced by topology analysis (bounding box, volume, hole counts, etc.) and the prompt content of the user conversation. It does not see your CAD geometry directly, and it does not have access to other customers' data. Detailed handling is described in our knowledge base article on AI analysis and user decisions.
Every co-pilot interaction is logged in your project. You can review what the assistant suggested and on which input. Outputs are reviewable by you — and by our internal QA team only with your consent.
Training and evals
The model is built on a third-party large language model with a layer of Uretyco-specific tools and prompt engineering. We run a regression test suite over a fixed set of representative parts every release. We do not train the underlying model on customer data; the only personalization is in your project's session memory, which you can clear at any time.
Where it helps most
Customers who are not domain experts (procurement officers, mechanical engineers from a different specialty, designers learning manufacturing) get the most value. Domain experts use the co-pilot mainly to compress the iteration cycle.
Where it fails honestly
Multi-part assemblies with intricate fits, parts with regulatory constraints (medical class III, aerospace AS9100), and exotic alloy/process combinations exceed the model's confidence. In those cases the platform routes the conversation to a human engineer. We will tell you upfront when this happens.
Related articles
Can't find what you need? Contact our team for expert assistance with your manufacturing project.