Machine Learning

The challenge - Why machine learning in CAD files?

Imagine starting a new job and your manager asks you to locate a file from 20 years ago that someone else created.

Where would you even start?

Physna has a unique process to classify and tag models based on their geometry.

The system will identify to its best ability and users will go and tweak the classification, if something doesn't belong, they'll take it out of that classification and Physna will take all the others automatically and wise versa, if something needs to be there and it's not, users will put it there and Physna will bring all the rest from other classifications.


My contribution - Core Classifications Experience

At the time I joined Physna, the backend was ready and the team has started building the front end. I met with the lead sales engineer to learn about this process and noticed that there are couple of things going on. Sales engineer knew Physna inside out, helping him solving some issues doesn't necessarily solves the customers issues.

The customers on the other hand, really saw the benefit of this ML but they didn't know how to re-classify in case something was off, they didn't want (or didn't care tbh how Physna goes about it) all they wanted was an easy way to tweak the predictions.

Prototype to validate the interaction

Implementation

Re-Classify Experience

After bringing the core UX to life, which was a crucial step in order to support the next iteration, I suggested we surface the classifications inside the Search results as another way of filtering (AKA Classification Match).

From here, the user was provided with either a link that will redirect them to the classification page to make tweaks, or they could just