Case Study: Learning Tools

Precision Needed

Software in the CAD (Computer-aided design) and CAM (Computer-aided manufacturing) space needs to be precise: when you are crafting a new vehicle or simulating the material performance characteristics of a coffee cup or an airplane wing, even the fraction of a fraction has to be as exact as possible. As much as it has advanced in the last few years and for all for their power, AI systems are not always know exactitude and precision: they have a tendency to hallucinate, invent new facts, come up with new reality — liberties you cannot afford in a context where every tiny part matters.

A learning co-pilot

With the need for accuracy in mind, our team worked with a leading CAM software developer to come up with a suite of learning and documentation tools, a co-pilot that would harnesses the power of AI without losing any degree of precision. It started by ingesting tens of thousands of specialty manuals and training documents including structured and unstructured data, storing them in CosmosDb, then creating embeddings and adding them a private Azure/OpenAI instance. This ensured the consistency of output while providing a secure environment to store proprietary data.

Pipeline and hybrid search

We also included a Cognitive Search and Cognitive Enrichment Pipeline to allow for new material to be continuously added, so that the model would evolve and grow organically over time. Searches were performed in a hybrid approach, maximizing the power of vector and semantic searches.

Smart UI

With an innovative and simple UI, users could type their queries and have them answered in multiple languages. We worked closely with teams of specialty manufacturing engineers and data scientists to ensure that users would get the most optimal and accurate answers in the language of their choice to facilitate a seamless workflow.

Contact us to learn how we can customize AI for you.