Character Lab

Enversed builds a lot of AI training experiences, like VR simulations where trainees talk to a virtual client or patient. Those AI characters used to be hardcoded in a build or scattered across Google Sheets, so changing a single trait meant a developer and a new build, with no version history or access control. Character Lab is a central platform where trainers configure characters, dialogues, environments and scenarios themselves, for several client organizations at once.

I started with research and wireframes, working out what a character needs to be configurable and how a non-technical trainer can move through it all without getting lost. A colleague turned my wireframes into the visual design, and I built the rest. The biggest decision was making everything schema-driven. Each type of content is described once on the server, and the editing forms build themselves from that description, so adding a field no longer needs a new deployment.

Characters come in Basic and Advanced modes, because not every trainer wants to engineer a personality. Basic sticks to the essentials, while Advanced adds archetypes, personality sliders and stress responses, pre-filled from the Basic profile. When two trainers started overwriting each other's work, I added realtime collaborative editing, where each field locks while someone is editing it. Trainers can also chat with a character to test it before it goes into a real training.

Character Lab is now a multi-tenant platform where companies build and manage their own AI characters and scenarios. It comes with roles and permissions, collaborative editing, test chats and usage analytics.

The platform is live and in active use, both internally and by client companies, and it feeds the characters into Enversed's VR trainings. It was also my first time taking a real product through its whole life, including versioning, data migrations, security and all the work that comes after the first version ships.