Product Design · Amazon Web Services · 2022 - 2025

AWS Project Libra

Libra is an AWS hybrid meeting room system built so remote and in-room participants can actually meet as equals. As the sole product designer, I owned the join flow, in-meeting experience, AI features, and a 400+ component design system. Shipped to AWS S-team leadership spaces from 2022 to 2025.

Role
Product Designer
Scope
End-to-end: join, in-meeting, exit
Timeline
2022 - 2025
Research
30+ validation studies
AWS Project Libra
13%
SCAT score improvement
(~$2.8M pipeline)
400+
Design system
components shipped
3,151+
In-room participants
across 477+ meetings

Business Context

Post-pandemic, AWS leadership saw a gap: video conferencing tools were built for fully-remote calls, not for rooms where some people sit together and others dial in. In practice, this created a two-tier meeting. In-room participants dominated conversation. Remote participants were invisible.

Libra was an S-team level bet to fix this. The system includes tablets at each seat, room TVs, and integrations with Zoom, Teams, and Chime. From 2022 to 2025, it was deployed across 477+ meetings with 3,151 participants.

"This is better than our OP1 executive rooms. Can we deploy in all leadership spaces?"

VP, AWS

"If I were doing a document review tomorrow, I'd want to do it in a Libra room."

Director of Workspaces, Amazon

The Problem

Hybrid meetings create two completely different experiences, and neither side is well served.

Remote participants are invisible

  • Can't read body language or see who's reacting
  • Can't tell who's speaking when voices overlap
  • Routinely talked over or forgotten mid-conversation

In-room participants have their own friction

  • Joining from a personal laptop creates echo for everyone
  • Sharing a single room mic means audio drops when the speaker turns away
  • Room layout forces everyone to face one direction, making natural conversation impossible

Libra's goal was to bring equity to the room: making hybrid meetings equally fair, clear, and comfortable for everyone, regardless of where they're sitting.

Design tenets that guided every decision:

Low touch Bring the user's focus back to the meeting itself. The system should never compete for attention. Every design choice aimed to reduce distraction, not add to it.
User trust People don't want to be filmed, tracked, or analyzed without control. Every AI and camera feature had to earn trust before asking for attention.
Surfaces Tablets at each seat, TVs on the wall, three third-party platforms. A decision that works on one surface can break another. Design once, adapt everywhere.

What I Delivered

High fidelity designs

End-to-end across join flow, in-meeting, and exit. Built responsive components and user flows for tablet and TV.

User research

40+ in-person usability tests. Ran large-scale surveys (N=491). Research directly shaped product strategy at the leadership level.

Accessibility

Designed for neurodiverse users, low vision, deaf and hard of hearing participants. Ran accessibility testing with real users. 100% of accessibility participants preferred Libra over traditional conference rooms.

Design system

400+ responsive components across all surfaces, defined the grid system. Single source of truth for design and engineering.

Meeting control interface TV display with AI meeting assistant

What Users Said

"

As someone in visual design that is often content sharing as a presenter, I can't tell you the number of times that people ask me to zoom in on one thing while another tells me to zoom elsewhere. With this pinch to zoom feature on individual tablets, meeting participants can clearly see and focus on their content.

"

The tablet feels more natural- it focuses the conversation more organically at the table and makes it feel easier to hear the remote people so it's more cohesive.

"

It's really cool. If I came to this room, I wouldn't need to bring my computer.

Deep Dive

AI Meeting Assistant

In early 2025, before AI meeting assistants became mainstream, we were among the first to ship one. I had 2 months, real privacy concerns from users, and no playbook for ambient AI in a multi-device meeting room.

2 mo
Kickoff to shipped beta
30%
Post-meeting time reduction
85+
NPS from initial pilot
AI Meeting Assistant floating overlay on meeting view

Research (N=129): The top user needs, meeting summarization (84%) and action items (83%), were both passive. Users wanted AI working in the background, not interrupting the meeting. This validated our Low Touch principle and anchored every design decision that followed.

Three Directions Explored

I explored three approaches. Each tested a different hypothesis about where AI belongs in a meeting room.

Option 1: Chat Panel Integration
Option 1
Chat Panel

Industry-standard sidebar. Low distraction, but felt like a chatbot, not ambient. Contradicted our low-touch tenet.

Option 2: Dedicated AI Tile Leadership push
Option 2
Dedicated AI Tile

Highly visible, but competed with video. Neurodiverse testing showed 40% higher cognitive load.

Option 3: Floating Insight Overlay Shipped
Option 3
Floating Insight Overlay

User-controlled, draggable, ambient. Gave people AI when they wanted it without competing for attention.

Have a Backbone
"AI needs to be prominent. That's our competitive edge." Leadership, pushing for Option 2 (Dedicated Tile)

I pushed back. Neurodiverse testing showed 40% higher cognitive load with the dedicated tile. One participant said, "I can't focus on both." The tile competed with the very thing users came to the meeting for: other people. I presented the evidence alongside our design principles and proposed Option 3. Leadership agreed.

Neurodiverse testing data 40% cognitive load increase "I can't focus on both" Low Touch principle

Final Design

After two rounds of iteration addressing clutter and video occlusion, the final version shipped to beta with catch-up insights, toggle controls, and seamless meeting navigation.

What I Learned

01
AI design is about boundaries, not features

The hardest decisions weren't what AI should do, but what it should not do. Staying silent is a design choice.

02
Data + user voice beats hierarchy

I didn't debate opinions. I brought user data and direct quotes. Evidence made the conversation collaborative.

03
Constraints force clarity

2 months, third-party dependencies, cross-platform. Every constraint forced a sharper decision.

Want the Full Story?

This covers the highlights. I'm happy to walk through the complete process, including research findings, design explorations, AI meeting assistant details, and key decisions, in a live conversation.

Request Full Case Study
Next project
LivePerson AI Career Site →
← All work