Virtual Agent Design of Rabindranath Tagore

A conversational AI designed to encourage children to learn about India’s late poets and authors.

A virtual agent is a conversational chatbot that uses AI and NLP to understand user intent, analyze patterns in a conversation and accordingly respond to users.

TIMELINE

MY ROLE

TEAM

TOOLS

1 month design exploration; implementation on going
(Summer 2022)

End-to-end design process from exploration, dialog flow, avatar and environment design, user test

Solo with guidance from professor Dr. Anirban Chowdhury (Prof. of Design UPES)

Figma
DialogFlow
Adobe Illustrator

PROBLEM SPACE

Children often learn about prominent literary figures of India through textbooks and rote learning.
This makes it an arduous task to engage with an author and their works.

Project proposed by Govt. of West Bengal for Nobel laureate and author Rabindranath Tagore.

USER GROUP: Children aged 10 - 15

INDUSTRY: Education / Ed-Tech

SOLUTION

A Minimalist and easy to navigate web-based platform with a virtual agent of Rabindranath Tagore. Focusing on conversation for learning.

Nobel Prize

Education

Childhood

Shantiniketan

National Anthem

Msg to younger generation

Conversation Prompts

Provide a way to start conversations on topics that children are most interested in and also provide educational value.

Error states and feedback

To communicate any system errors to the user.

Engaging animations and sounds

Make the experience more immersive and interactive enhancing the user experience.

Hover over links to know more about a topic or word

Allows users to understand the meaning of a word or read about a topic without having to look it up.

Basic settings for a conversational AI

Manage microphone, speaker, volume and language settings

And here's how we got there:

understand

design

develop

validate

Interviews, empathy maps, affinity mapping

Exploring avatar expressions & colors, conversation prompts design

Voice script, dialog flow

User testing, design system

01 UNDERSTAND

User Empathy Map

Conducted interviews with children aged 10 - 15 and evaluated their requirements when learning about authors & also when interacting with a virtual agent. This helped reveal clear opportunities to focus on in the CUI. Patterns emerged on - what they wanted to learn from the author, the appearance of the agent, ease of use and satisfaction.

02 DESIGN

Dodging the Uncanny Valley

The Uncanny Valley hypothesis is based on the observation that as robots appear more human like they appear more appealing - but only up to a certain point. Upon reaching the uncanny valley, our affinity descends into a feeling of unease and tendency to be scared or freaked out. I kept this in mind while designing Tagore's avatar, and designed it to be “cartoon-like” which is suitable to our user group.

Exploring Avatar Expressions

Drawing from the empathy map, I identified the key expressions an avatar will need to display that best empathise with the users. This ensures feedback to the children and more expressive conversation.

I finalised 4 basic expressions of the avatar -
thinking (initial expression), listening, incomprehension and happy.

Thinking

Listening

Incomprehension

Happy

Exploring Conversation Prompts

Affinity mapping from user interviews led to a group of topics that users were interested in learning about. These topics were made into icons/buttons that serve as prompts for conversation.

Education

Childhood

Nobel Prize

Shantiniketan

National Anthem

Message to the younger generation

03 DEVELOP

DialogFlow's Elements of Voice Design

DialogFlow is a Google-owned framework that enables users to develop conversational applications that support Natural Language Processing (NLP)– such as chatbots and virtual agents.

INTENT

INTENT MATCHING

ENTITY TYPE

RESPONSE

The words the user speaks.

The information the user wishes to gather.

Matching intent to
end-user expression.

user expression

Information the app needs to complete the task.

The output from the virtual agent.

"Can you tell me about your childhood?"

To know about childhood.

School, play, activities...

"I was the youngest of 13 children....."

Sample Script in Action

This is a sample script for the intent Nobel prize. Here the user wishes to gather details about Tagore and his winning the Nobel prize. Care was taken into forming the simple sentence structure and positive tone of the language. Questions are framed to further encourage children to respond.

Try the chatbot using this Link

04 VALIDATE

Usability Testing

After building a concept prototype a thorough testing of the dialogs as well as UI accessibility was done using a live dialogue method
(n=3 , children). After iteration, users were able to communicate more naturally and cohesively with the virtual agent.

Defining the Design System

TYPOGRAPHY & COLOURS

REFLECTION

Keep it simple

Consider error strategy

Iterate iterate iterate

Especially designing for children, it’s important for the conversation flow to feel simple and natural, yet exciting. When designed correctly our virtual agent can induce a feeling of fun while learning and ease of use.

When designing the script, we should always design for any incomprehension scenario where the assistant does not understand.

A successful design process is collaborative with continued iterations from feedback. To leverage visuals alongside voice interaction, to create rich conversational experiences especially for children, I learned to test the screens and iterate accordingly.