While at Kentauros, I developed Scout, a platform for training and deploying computer user agents.
As we we developed our own computer-use agents it became clear that we needed to be able to grade agents task attempts, so Scout was developed to address this. The goal of the platform is to generate training data that can be used to improve user defined skills and eventually run those tasks reliably. I took the lead on the design and frontend architecture of the platform, eventually shipping it to our industry partners, who celebrated the UX and ease of use.

Individual agent actions can be reviewed afterwards, allowing fine grained feedback data to be generated.