Bondmate : An AI Companion

Bondmate : An AI Companion

Bondmate

EVENT

EVENT

AWS/Meta Hackathon at Amazons San Jose Office

AWS/Meta Hackathon at Amazons San Jose Office

TEAM

TEAM

Khang Nguyen (VR Game Developer), Michael Lee (VR Game Developer) Tarun Hari (Product Designer/Manger) Meuy Xaimuangmonh (UX/UI Designer)

Khang Nguyen (VR Game Developer), Michael Lee (VR Game Developer) Tarun Hari (Product Designer/Manger) Meuy Xaimuangmonh (UX/UI Designer)

MY ROLE

MY ROLE

AI Product Designer & Manager

AI Product Designer & Manager

MY CONTRIBUTION

MY CONTRIBUTION

LLM Prompting (Bedrock), WebSocket Integration, Spatial UX, Prototyping, Product Management

LLM Prompting (Bedrock), WebSocket Integration, Spatial UX, Prototyping, Product Management

The Problem: Digital companions are often "trapped" within a single device or interface, creating a fragmented emotional experience as users move between 2D screens and 3D environments.


The Challenge: How can we design a persistent AI entity that maintains its persona and context across a seamless transition from a web dashboard to a Meta Quest 3 headset?

The Problem: Digital companions are often "trapped" within a single device or interface, creating a fragmented emotional experience as users move between 2D screens and 3D environments.

The Challenge: How can we design a persistent AI entity that maintains its persona and context across a seamless transition from a web dashboard to a Meta Quest 3 headset?

The companion addresses isolation and loneliness by fostering social interaction and enhancing team connections. Bondmate also helps overcome communication barriers, ensuring everyone feels included and minimizing miscommunication. By boosting motivation and productivity, it aids in maintaining focus and engagement.


Additionally, Bondmate supports mental health by reducing stress, alleviating anxiety, and lowering the risk of burnout. It also promotes better work-life boundaries, helping users disconnect and maintain a healthy balance between their professional and personal lives.

The companion addresses isolation and loneliness by fostering social interaction and enhancing team connections. Bondmate also helps overcome communication barriers, ensuring everyone feels included and minimizing miscommunication. By boosting motivation and productivity, it aids in maintaining focus and engagement.

Additionally, Bondmate supports mental health by reducing stress, alleviating anxiety, and lowering the risk of burnout. It also promotes better work-life boundaries, helping users disconnect and maintain a healthy balance between their professional and personal lives.

LLM Configuration

LLM Configuration

The goal of the hackathon was to use Amazon Bedrock to build the Agent. I defined the personality of the Agent by tailoring it for remote workers by making it empathetic and caring


The goal of the hackathon was to use Amazon Bedrock to build the Agent. I defined the personality of the Agent by tailoring it for remote workers by making it empathetic and caring


Cross-Platform Logic

Cross-Platform Logic

The core challenge was preventing "context drop" during device switching. By utilizing a central WebSocket, the AI companion recognizes the user's active platform in real-time. This architecture allows the companion to shift from a text-heavy assistant on the React dashboard to a fully expressive, spatial entity in Unity without losing a single beat of the conversation.


The core challenge was preventing "context drop" during device switching. By utilizing a central WebSocket, the AI companion recognizes the user's active platform in real-time. This architecture allows the companion to shift from a text-heavy assistant on the React dashboard to a fully expressive, spatial entity in Unity without losing a single beat of the conversation.

Reflection

Reflection

As a Spatial AI Design Technologist, my goal for Bondmate was to move beyond the 'chatbot' and create a persistent agent that inhabits the user's environment. Utilizing Amazon Bedrock, I engineered a multi-modal persona capable of maintaining context as users transitioned from 2D web dashboards to immersive Meta Quest 3 spaces.

This project was a deep dive into behavioral architecture: fine-tuning LLM system prompts to ensure the AI felt consistent, whether it was a text-based productivity partner or a spatial companion. I solved for state-persistence by building a WebSocket-based handoff, ensuring zero latency in conversation memory during device switching. This experience solidified my expertise in balancing high-level agentic logic with the technical constraints of real-time spatial computing.

Interested in working together? Let’s collaborate!

Interested in working together? Let’s collaborate!

Interested in working together? Let’s collaborate!

© Tarun hari 2026