Answering specialists in-context — an AI support assistant that meets users at the moment they're stuck, instead of routing them into a support queue.
Vault RIM is a dense regulatory platform used by specialists working under compliance pressure. The complexity meant users regularly hit questions the interface couldn't answer on its own — and every unanswered question became a support ticket. The support team absorbed the cost of the platform's learning curve, and users lost time waiting for answers to questions that were, in principle, already documented somewhere.
The opportunity: answer users in-context, at the moment they're stuck, instead of routing them out of the platform and into a queue. The constraint: this is a regulated environment, where a confidently wrong answer is worse than no answer at all.
Designed and built a proof of concept for an AI support assistant on Novo Nordisk's ChatBrd framework — from concept through to a working prototype that could be tested with real users and real questions. The design work was as much about boundaries as capability: what the assistant should answer, what it should defer, how it signals uncertainty, and how it stays anchored to source documentation rather than improvising.
The POC was validated, taken into production, and released to end users and support teams in June 2025 as RIMBot.