Zenith VR · interaction design · 2026
What if an AI moderator knew when not to speak?
An academic VR concept exploring how students can join conversations with less pressure - and how AI can help without interrupting.
Presence first. Assistance when you choose.
Role
Interaction Designer
Research synthesis, interaction design, prototyping, and evaluation in a four-person team.
Timeline
First Semester
MS HCI core course · Interaction Design Practices
Type
Academic Case Study
Research-led VR concept · Not shipped
Output
VR Prototype
Unity · Meta Quest / Quest Pro
01 / Problem & research
Students needed less pressure, not more prompts.
Five undergraduate and graduate student interviews revealed forced study sessions, post-class isolation, uneven participation, and awkward silence. Literature on group facilitation supported the direction: make it easier to enter a conversation without demanding attention.
From friction to design requirements
Select a theme to explore the design response.
Arrive before you contribute.
Let students join and settle before asking them to speak. Presence should be a valid starting point.
02 / Design decision
Let conversation emerge before guiding it.
We compared active AI facilitation with a quieter shared space. Passive presence better matched the need for low-pressure connection, even if conversation took longer to start.
The people lead.
The room supports presence first. AI stays optional, at the edge of the conversation.
Concept B was the team’s selected direction.Concept A · AI-facilitated collaboration
- Automatic group matching
- Structured prompts
- Guided discussion flow
Fast interaction, but higher social pressure.
Selected · Passive presence + subtle facilitation
- Join without speaking
- Interact through proximity
- AI stays in the background
Lower pressure and closer to organic in-person interaction.
A circular layout gives everyone equal visibility. A visible AI toggle keeps help optional. The assistant should offer an opening, then step back - not manage the conversation.
03 / Prototype
Build the room. Simulate the moderator.
We prototyped the journey from arrival to contribution in Unity. Manually triggered AI assistance let us test the social interaction before investing in automation.
One room. Five moments.
A clickable visual explanation - not the Unity prototype.
Lobby
Orient, learn the controls, and join without interrupting the room.
Join at your own pace.
Look around. Learn the controls. Enter when ready.
- Built in UnityA spatial lobby and circular room for Meta Quest / Quest Pro, with joystick movement and controller interaction.
- Explored input pathsVoice, virtual keyboard, and talk-to-text explored different comfort levels. Testing still exposed friction with movement and text entry.
- Simulated AIIn a Wizard-of-Oz setup, the team manually triggered prompts to test timing, usefulness, and trust. This was not an autonomous AI system.
The flow above explains the intended experience; it is not a recording of the Unity prototype.
04 / Testing & findings
The room helped. The interruptions did not.
Four external evaluators tried three scenarios: two in a headset and two through a guided remote walkthrough. Feedback supported the shared-room concept, but exposed problems with AI timing, text entry, and discoverability.
Join and navigate the lobby; hold a group conversation with AI prompts; brainstorm three ideas and capture them in the shared workspace.
Observed friction → proposed next iteration. These changes have not been validated in a follow-up study.
Priority 01 · CriticalMake AI legible and well-timed
Prompts sometimes arrived during natural pauses. Evaluators could not tell why the assistant had intervened.
Proposed next iteration- Wait longer before prompting
- Show a cue before the AI speaks
- Explain the trigger in one line
- Let people dismiss or postpone
A prompt during a pause.
An unexplained interruption can make a natural pause feel like a problem.
Priority 02Reduce input friction
Joystick navigation and the virtual keyboard increased the cost of participation for first-time users.
Proposed next iteration- Add voice-to-text
- Support quick reactions
- Consider physical keyboard pass-through
Text entry takes effort.
Navigation and virtual typing add effort for first-time participants.
Priority 03Strengthen onboarding
Collaboration tools existed, but some were not discoverable without guidance.
Proposed next iteration- Teach movement in the lobby
- Highlight tools on first use
- Keep a persistent help cue
Tools without a clear introduction.
Useful collaboration tools can remain unnoticed without first-use guidance.
Feedback highlighted immersion, equal visibility, and a more enjoyable sense of collaboration. Well-timed prompts could help participation. These were qualitative signals, not proof of better meeting outcomes.
Four evaluators is a small sample, and remote participants could not assess motor comfort or headset navigation. A larger, all-in-headset study would be needed before making outcome claims.
05 / Reflection
Helpful AI needs restraint - and a clear reason to act.
I learned to prototype the social consequence, not just the interface. Comfort without clear controls creates hesitation; an unexplained prompt becomes another interruption. Zenith remains an unshipped academic prototype, not evidence that AI improves meeting outcomes.
Refine prompt timing, explain why help appears, and simplify first-use controls. Then run a larger, all-in-headset study before making claims about participation or meeting quality.
Designing with the team.
I contributed to concept and interaction design, research synthesis, prototyping, and evaluation. The work was collaborative; “we” refers to the team, not sole authorship.
01Joel Akinde
02Shashank KatariyaPortfolio author
03Cory Savin
04Kiersten Foulk
Next case study · 01