The Problem
greytHR’s virtual assistant Bella was live, but she wasn’t consistent.
Sometimes warm, sometimes clinical, never quite the same twice. For an HR platform where employees interact with Bella during sensitive moments (like leave requests, payroll queries, and onboarding), inconsistency is more than just a UX problem.
The design team had already done foundational personality research and landed on an INFJ profile for Bella. She was thoughtful, empathetic, and quietly confident.
My job? Taking this framework and making it real inside every line of dialogue she spoke.
Bella 🤖
Conversational AI Design for Greyptip Software.
Role: UX Writer & Conversational Designer.
Testing and Iteration
We pushed dialogue to Greytip staff for testing early and often. The feedback was consistent: The language worked, but the tone needed some tightening. That became the central tension of hte project. How do you maintain warmth and consistency at scale, across hundreds of dialogue states, written by more than one person?
The answer was documentation. Building voice guidelines became a shared contract between writers, designers, and engineers. Anyone coming across Bella’s language design needed a north star, and we built one.
What I Did
Joining mid-process, I worked alongside an associate UX Writer in close collaboration with the design lead and cross-functional stakeholders. Our weekly rhythm involved standups, design reviews, and regular check-ins to pressure-test whether the words on screen matched the product vision.
My core work involved building out Bella’s conversation flows. I mapped what she says, when she says it, and how her responses shift depending on input. Empathy had to be more than just a design principle!
The Outcome
Bella shipped! She’s live today on the greytHR app and website, interacting with HR teams and employees across India. The voice guidelines we created during this project gave a better standard for any future conversational design work.