• 

  NAOMI quiet home screen and low-pressure interaction entry points.

• 

  Human-adaptive listening demo with pressure-controlled responses.

• 

  State organization and staff note view for non-diagnostic handoff.

• 

  Runtime integration debug panel showing Gemini, Agent Engine, Arize MCP, and Trace ID.

Inspiration

Many AI systems focus on giving answers as quickly as possible. But people are not always ready for answers.

Sometimes they are tired. Sometimes overwhelmed. Sometimes anxious. Sometimes they simply need help organizing their thoughts.

In those moments, immediate advice can increase pressure rather than reduce it.

However, in healthcare, caregiving, welfare, and daily life, people are often tired, overwhelmed, anxious, or unable to clearly explain what they need.

I wanted to explore a different approach:

Instead of immediately providing answers, what if an AI could first understand a person's state and adjust how it interacts with them?

This idea became NAOMI.

NAOMI stands for:

Natural Adaptive Observation & Meaningful Interaction

A Human-Adaptive Listening Agent designed to reduce conversation pressure and help people organize their thoughts before advice is given.

What it does

NAOMI observes a user's state and changes its communication style accordingly.

It focuses on:

• Listening First
• Pressure Control
• Cognitive Load Reduction
• State Organization
• Supporting caregivers and support staff

Our goal is not to replace professionals.

Our goal is to help people communicate more clearly with professionals, caregivers, family members, and support staff.

Challenges

One challenge was avoiding the typical AI behavior of immediately providing solutions.

Another challenge was creating a system that feels supportive without becoming overly intrusive or acting like a medical diagnosis tool.

The project continues to evolve through user feedback and real-world testing.

We believe the future of AI is not only about becoming smarter.

It is about becoming more human-aware.

NAOMI explores a future where AI adapts to people, instead of forcing people to adapt to AI.

NAOMI uses the required hackathon runtime integrations:

• Gemini is called at runtime in agent/gemini_brain.py through generate_content().
• Google Cloud Agent Engine is called at runtime through agent/agent_engine_client.py.
• Arize Phoenix MCP Server is called at runtime through ClientSession.list_tools() and ClientSession.call_tool(...).
• The deployed Streamlit app includes a debug expander showing Gemini / Agent Engine / Arize MCP runtime status and a trace ID for judges.

※The first access may take 20–30 seconds because the Cloud Run container can cold start after inactivity.

Built With

• accessibility
• agents
• ai
• arize
• cloudrun
• design
• engineering
• gemini
• generative
• googiecloud
• google
• human-adaptive
• mcp
• opentelemetry
• phoenix
• prompt
• python
• streamlit
• ux