Adaptive Tone & Voice for an AI Agent

9:41

Jane Doe

@username

How are you feeling today?

3

Connections

3

Invites

Bonding

Social

Creative

Knowledge Collector

Explorer

Curious

Dreamer

Train you agent

Description of the section

What inspires your creativity

Answer

When do you feel most energized?

Early mornings

Midday

Evenings

It depends on the day

Home

Rooms

Summary

Odyssey is an AI social matching app that creates a complex vector representation of the user’s personality and interests to match them up with people within the same “cluster,” very similarly to how AI infers meaning through proximity of objects in the vector space. The more time the user spent within the app, the better it knew them, the better connections it could recommend. I was brought in to shape the app agent’s evolving tone and voice framework as well as contribute to the system prompt.

Client: Odyssey

Role: Lead UX Designer

Activities:

  • Workshop Creation + Facilitation
  • Voice & Tone Guide
  • Conversation Design
  • Prompt Engineering

Process Details:

  1. There was a famous leak of system prompts from major products. I took this opportunity to study how the best prompts are structured, how tone & voice sections and examples were phrased, and the specificity and breadth of guardrails to come with general template for an effective system prompt.
  2. Created a gamestorming workshop and guided the client’s team in the exploration of the Odyssey’s “default” agent voice, which currently was devoid on brand or personality, as well as the adaptive framework of how communication style could change as the agent “bonding level” grew over time.
  3. Using the findings, I built out tone & voice guidelines for the default agent voice and come up with the 4-bonding levels framework for how the voice and style changes as the agent becomes more familiar with the user.
  4. I converted the tone & voice guidelines into a system prompt, and started interacting with and evolving the prompt in a sandbox environment, stress testing and adding necessary guardrails.

Workshop Activites

Prompt Engineering

Adaptive Tone & Voice for an AI Agent

9:41

Jane Doe

@username

How are you feeling today?

3

Connections

3

Invites

Bonding

Social

Creative

Knowledge Collector

Explorer

Curious

Dreamer

Train you agent

Description of the section

What inspires your creativity

Answer

When do you feel most energized?

Early mornings

Midday

Evenings

It depends on the day

Home

Rooms

Summary

Odyssey is an AI social matching app that creates a complex vector representation of the user’s personality and interests to match them up with people within the same “cluster,” very similarly to how AI infers meaning through proximity of objects in the vector space. The more time the user spent within the app, the better it knew them, the better connections it could recommend. I was brought in to shape the app agent’s evolving tone and voice framework as well as contribute to the system prompt.

Client: Odyssey

Role: Lead UX Designer

Activities:

  • Workshop Creation + Facilitation
  • Voice & Tone Guide
  • Conversation Design
  • Prompt Engineering

Process Details:

  1. There was a famous leak of system prompts from major products. I took this opportunity to study how the best prompts are structured, how tone & voice sections and examples were phrased, and the specificity and breadth of guardrails to come with general template for an effective system prompt.
  2. Created a gamestorming workshop and guided the client’s team in the exploration of the Odyssey’s “default” agent voice, which currently was devoid on brand or personality, as well as the adaptive framework of how communication style could change as the agent “bonding level” grew over time.
  3. Using the findings, I built out tone & voice guidelines for the default agent voice and come up with the 4-bonding levels framework for how the voice and style changes as the agent becomes more familiar with the user.
  4. I converted the tone & voice guidelines into a system prompt, and started interacting with and evolving the prompt in a sandbox environment, stress testing and adding necessary guardrails.

Workshop Activites

Prompt Engineering