We caught up with Jellypipe, and learned they are doing some interesting things with AI.
The company has been around since 2017, and operates a 3D print network. In this model, requests from buyers are routed to an appropriate provider, which executes the print and ships the results.
That’s quite similar to how several other 3D print networks operate, but Jellypipe has an added twist: they realize that many prints can fail because they are not the correct geometry, or require peculiar print settings. To avoid these issues, their service has been bundling a consultant with the print requests to ensure things will turn out OK. That’s quite unlike other print services where you can have the “garbage in, garbage out” problem.
They’ve had quite a bit of success with this, model, but it seems they’re adding some new function.
During a recent encounter, they explained to us that they’ve completely rebuilt their platform, and that it now (at least in beta) includes a new AI assistant.
How does it work? The assistant responds to queries from customers and offers advice. One feature is the ability to tell the customer which materials would be optimal for the application. For example, a part intended for food use might be provided with the suggestion to use an oil-resistant material.
The assistant is also able to create dynamic price quotes for parts production based on the information provided.
I asked where the assistant gets the information for its analysis, and was told that they don’t simply load data from the open web, but have their own internal materials database.
The assistant also learns as it processes requests, and gradually gets better.
Jellypipe built the new assistant on top of OpenAI’s GPT4 LLM system, in a manner similar to that done by many other chatbots. One important side effect benefit of this approach is that GPT4 understands a wide range of languages. This means that it would be possible for Jellypipe’s assistant to be easily used in many regions.
The assistant is still in beta, being tested by a trusted few. As you might imagine, there is a lot of testing required to make sure the service works properly. Then it will be launched first to their solution partners, and finally to customers in general. They’re expecting this to occur some time this month.