Chat ADM — Internal AI Assistant (Alpha)

Chat ADM is a secure, internal AI assistant for teams to ask questions, summarize files, and generate draft content with company context. I led UX from discovery to high-fidelity prototypes, focusing on trust, clarity, and guardrails—including a Code of Conduct banner, visible limits, and transparent system messages. The result is a calm, familiar chat UI that reduces friction and earns confidence.

Understanding the project landscape

Why generic chat UIs fall short for enterprise

Most AI chat tools prioritize speed over governance and clarity. Internal users need a space that’s familiar and safe: clear rules, visible limits, reliable context, and predictable outputs. We set out to design a trust-first chat that balances power with guardrails and transparency—so teams can move quickly without second-guessing what’s happening.

Research and discovery phase

Ground truth from real usage and prior insights

Audit of prior interviews: Consolidated findings from a previous designer; identified gaps around file upload, traceability, and data safety.

Survey (n=10): If file upload were available, Daily (5), Weekly (4), Other (1) → clear demand for document analysis & summarization.

Top use cases: Summarize large docs; extract action items; analyze content; draft emails and reports; “teach me” queries.

Trust & governance: Users want a visible Code of Conduct, source links, and explanations for limits (e.g., chat cap, token limits).

IA needs: Fast navigation to recent chats, projects/workspaces, and a clear “New Chat” affordance.

Accessibility: Keyboard-friendly composer, strong focus states, AA contrast in both Light & Dark modes.

‍

Initial design exploration

Familiar patterns, enterprise guardrails

Early wires leaned into a calm chat metaphor with clear structure and safety cues.

Sidebar: Recent chats, project folders, and a large New Chat button.
Composer: Attachments, Persona selector (system prompts), and a visible character counter.
System guardrails: Persistent Code of Conduct banner; chat-limit notice; update card component.
First-run: Cards suggest common tasks (ask complex questions, share feedback) to reduce blank-page anxiety.

‍

Refining the design approach

Reduce cognitive load, surface traceability

Iterations emphasized scannability and explainability.

Message anatomy: Clearly separated user vs. assistant blocks; inline citations and “copied” feedback.
Context clarity: Thread name, system prompt badge, and last updated metadata within reach.
Theme polish: Harmonized Light/Dark tokens; ensured AA contrast; consistent elevation & spacing scale.
Empty states: Task cards tuned to most frequent actions; microcopy revised for warm, directive tone.

‍

Final design deliverables and key features

The experience, distilled

Trust-first UI: Code of Conduct banner, visible chat caps, and update announcements.

Calm chat model: Familiar layout with sidebar, thread titles, and quick New Chat.

Smart composer: Persona presets, attachments, token/length hints, and keyboard shortcuts.

Explainable answers: Inline citations (when available), copy action, and contextual footers.

Light/Dark design system: AA-compliant color tokens, components, and elevation rules.

Design ops: Reusable templates for cards, messages, alerts, and settings.

‍

Project impact and user feedback

Early signals from pilot

In pilot, usage and feedback pointed to a strong product-market fit for internal workflows.
‍

✅ Clear demand for file upload & document QA (5/10 daily, 4/10 weekly interest)
✅ Users praised guardrails (Code of Conduct / update notices) and the calm, predictable layout
✅ Dark & light themes improved accessibility and adoption across teams
✅ Research cadence + system components accelerated iteration and stakeholder buy-in

Explore more design work

Interested in seeing how design can transform digital experiences? Check out more case studies showcasing innovative UX solutions

Projects Contact