- Role
- Senior UX Designer
- Tools
- Figma, OpenAI
Overview
Context
After first-wave GPT instrumentation, the organization needed a branded, enterprise-ready chat interface; room to add models and modalities; and a flexible pattern that could evolve on a regular cadence. Leadership treated it as a foundational internal product—not a throwaway pilot.
Scope included competitive review, flows, low-fidelity exploration, branding and style guide, and 35+ high-fidelity desktop screens plus mobile treatments, with a prototype for stakeholder validation.
Challenge
What needed to change
The core question was how to combine multiple AI capabilities—standard and reasoning models, agents, and external ecosystems—without overwhelming users or locking the UI into a short-lived pattern.
- Unify chat, model selection, data sources, and app switching in one coherent shell
- Introduce a dynamic context pane that updates with the conversation
- Stay within brand guidelines while establishing patterns other teams could reuse
- Deliver at sprint speed with a small design team
Goals
Design principles
- 1. Establish a reusable layout and design system extension for internal AI initiatives.
- 2. Make multi-model and multi-modal selection legible with clear states and progressive disclosure.
- 3. Design the dynamic pane as a productivity surface—not only a transcript.
- 4. Ship desktop and mobile experiences that could scale with the roadmap.
Chapter 1
Design system and brand language
Color, typography, and component rules were defined so high-fidelity work stayed consistent as the product grew—and so other internal AI initiatives could align to the same reference.
Chapter 2
Desktop experience
Light and dark treatments for the primary workspace show how the same layout adapts across theme preferences while keeping hierarchy and pane structure stable.
Chapter 3
Mobile entry and responses
Mobile flows cover landing, navigation, and response states—keeping parity with the desktop mental model on smaller viewports.
Outcomes
Impact
- ↑ Delivered 35+ high-fidelity desktop + mobile screens within a 10-week sprint cycle
- ↑ Established reusable core layout adopted by multiple internal AI initiatives
- ↑ Designed scalable multi-model, multi-modal selections
- ↓ Reduced friction in internal knowledge retrieval across policy, HR, and product systems
- ↑ Positioned the organization for biannual AI interface evolution
- ↑ Design system became the foundation other teams referenced for AI-enabled tools
The work centralized AI experimentation into one governed interface, reduced time spent hunting documentation, and gave leadership a reference model for continued iteration.
Reflection
Closing thoughts
Enterprise AI adoption fails when interfaces are fragmented or model selection is opaque. This project balanced future flexibility with a clear, reusable layout—so teams could ship fast without painting the product into a corner.