Embodied Interfaces Studio
2026
Votive: AI in environments
A candle that listens to the room and burns according to what's being said.

Overview
AI Conversation Feedback + Reflection Device
This tabletop device is a creative research object exploring the imbalance in romantic conversation. It was inspired by common dating app experiences, particularly on Hinge, where people often leave dates feeling as if they spent the evening interviewing someone rather than engaging in an equal exchange. Votive is a physical embodiment of AI that gives an honest reflection, qualitative and quantitative data, between 2 people, of the conversation in print.
The device listens only to the presence of voices, not their content. At the conclusion, it prints a small receipt showing the total duration of the interaction and the percentage of time each person spoke.
By turning conversational dynamics into a quiet, physical artifact, the project invites reflection on listening, reciprocity, and the distribution of attention during intimate encounters without assigning judgment or scores.
Problem
Young people are dating less and have built a culture of not sharing critical feedback
Votive begins with a simple observation: people in their twenties have access to more ways to connect with each other than any generation before them, and yet the quality of those connections keeps getting worse. Dating apps have turned the early stages of relationships into a sorting exercise. Swipe, match, text, ghost. Optimized for volume rather than depth. When two people do end up in the same room together, they can often leave with two different perspectives of the event. and due to delays in developing social skills, good conversation can rarely happen. The muscle for genuine conversation, for asking real questions and sitting with uncomfortable answers, has atrophied from disuse.
57%
of single Gen Z adults aged 18–29 are not actively seeking a relationship or even casual dates — the highest rate ever recorded for this age group.
80%
of Gen Z reported feeling lonely in the past year, compared to just 45% of Baby Boomers. Gen Z is the loneliest generation ever recorded.
26%
of Gen Z adults aged 23–29 are married, compared to 42% of Millennials at the same age. Marriage rate has dropped from 6.1 to 4.2 per 1,000 people since 2019.
53%
of adults under 30 have used a dating app — the highest rate of any age group
Design
Prototyping AI's presence in objects
Targeted Environment
First dates and meetings can occur in a variety of different spaces and contexts; for this project, I wanted to focus on a two people who are seated at a small tabletop. Lower lighting that is common for evening, romantic experiences.

3D Sketching: Focus on the experience and functionality over appearance
Form 1 is designed to be a functional model for that moment when two people are physically present with each other but the conversation stays on the surface, circling familiar territory, never quite landing anywhere that matters. It places an object between them that listens without judgment, makes the invisible dynamics of the conversation visible, and at the end produces a document that says: this is what happened between you.


1. Form 1 features breakdown 2. Hardware and Software input/output diagram
Form 1 User Experience

The Output: Conversation Receipts
The purpose of Votive's receipt function is to give a record and summary of the conversation had between the two participants, providing a third-party perspective on the history of the interaction. Version 1 was the first tested receipt, which had more emphasis on AI's description of what happened. Version 2 was created following feedback to focus more on the conversation itself rather than the topics discussed.

Receipt output editing
Proof of Concept
Form 1 was a prototype to test initial assumptions about people’s interest in visualizations and AI-generated summaries of human conversations.

Ideation
Improving the model
Form 2: Model redesign
Votive is designed for that moment when two people are physically present with each other but the conversation stays on the surface, circling familiar territory, never quite landing anywhere that matters. It places an object between them that listens without judgment, makes the invisible dynamics of the conversation visible, and at the end produces a document that says: this is what happened between you.


1. Form 2 features breakdown 2. Hardware and Software input/output diagram
Dynamic states: Lighting
Feedback from the proof of concept indicated an opportunity for real-time data analysis. Different light configurations were designed to represent conversational qualities and tested with users to refine them.
Receipt takeaways: AI interpretation
For demo day, I planned to test new information displayed on the receipt as well as provide a way for participants to submit their experience in a survey they could access through a QR code on the receipt. For content, programmed the AI to display the percentage of conversation participation between the two participants, the question count, the energy level of the conversation, and a shorter, more grounded summary of the conversation made by ChatGPT.

Testing
Learning from observation and collecting data
Test 2: Playtest Demo Day
Playtest Demo Day served as an opportunity to place several people with different levels of familiarity with the project and the other participants, allowing for rapid prototyping of aspects of the environment and Votive's presence over 4 hours.
Purpose: Staging moments for conversation between two people for feedback on the Votive object and experience Goal: Capture feedback on what works and what doesn’t work to inform later iterations

Feedback and Observations from Demo Day


1 Q: What about this experience worked well?; 2 Q: What about the experience could have been better?, 3 Q: How was the experience with the Votive during the conversation?, 4 Q: How would you describe the quality of this conversation compared to a typical conversation with this person?, 5 Q: Upon arrival, what did you make of the object?, 6 Q: How would you describe the relationship you have with the person you were speaking with?, 7 Q: Did the Votive's presence affect the conversation?, 8 Q: Did you say anything during the conversation you would not have said otherwise?, 9 Q: Who do you think spoke more during this session?, 10 Q: How did you find the instructions?
Feedback
Improve the responsiveness of the lighting, making it reflect the status of the conversation in real time
Light, “fire” responds to the conversation in 5-second intervals based on volume, amount of conversation, and which microphone is receiving more input
Receipts are fun to explore, but some of the information is not important to the participants
Limit quantitative data to how much each person spoke, the number of questions asked, and include overall observations, as well as the AI’s guess to what kind of relationship the participants appear to have
The majority of groups hesitated after pressing the button
The LED of the button turns on after pressing, and the fire dances to signal that the session has started and the AI is recording
The interaction between the participants and the Votive was limited
After 2 minutes of conversation, have Votive print a question based on the discussion so far that could help stimulate the interaction
Test 3: Furthering the experience
For the final review, I had new users test the edited experience of Votive, this time including an ability for the device to ask a question after a determined amount of time, a change in the receipt metrics output, and improved the physical interaction with Votive by fixing the button.
Purpose: In a real-time test, the final iteration of this form will get feedback from participants about how Votive impacted their conversation and perspective on their relationship Goal: Test AI interaction assumptions and capture feedback on what does and does not work for later iterations

Interaction with the AI
The interaction between the participants and the Votive was previously only retrospective, with a large portion of the receipt dedicated to quantitative data. The interaction with the AI would only occur at the end of the meeting, which some found captured the conversation well in an artifact, but they did not feel the LLM was helping them improve their conversation beyond delivering stats and lights that reflected the meeting's tone.

Additional communication form tests
For one test, we ran out of paper, so we moved to “printing” the question slips and receipts on a dedicated browser page that connected to the Raspberry Pi. This was a fast way to convey the changes made to Votive without compromising the participants’ experience with the data collected.

Outcome
Making the invisible, visible
Three rounds of testing taught us something we didn't expect: the object matters less than the situation it creates. The menu questions gave people permission to ask things they already wanted to ask. The fire gave them something to look at when a question landed somewhere uncomfortable. The receipt gave the conversation a weight it wouldn't have had otherwise. What we learned about directing the AI output was really about directing attention: the more specific the data the receipt could reference, the more people felt genuinely observed rather than processed. The next step is removing the scaffolding. A first date in a more natural setting means the menu disappears, the instructions disappear, and the object has to do more of the work on its own. That is a harder design problem and a more interesting one.



