Dew
An AI learning assisstant
A magical AI companion for streamlined tasks, enhanced productivity, and active learning.
Duration: ~1 month
Product Design
UX Research
Character Design
Dew
An AI learning assisstant
A magical AI companion for streamlined tasks, enhanced productivity, and active learning.
Duration: ~1 month
Introduction
Introduction
Picture a bustling workplace where the daily grind is often hindered by the challenges of knowledge sharing, seeking expertise, and managing technical tasks. In this dynamic environment, efficiency takes a hit as employees grapple with the complexities of day-to-day tasks while juggling multiple software applications. The quest to find solutions to these persistent issues led to the creation of something remarkable - a solution that aimed to revolutionise the way we work and learn.
This case study is a journey into creating a user-friendly AI Work/Learning Assistant. In this case study, we'll go through the hurdles my team found in sharing knowledge at workplaces and how Dew came to the rescue. You'll see how Dew evolved from an idea to a mascot and a solution that seamlessly fits into ones daily tasks, helping them learn and work better, and paving the way for knowledge-led growth.
Picture a bustling workplace where the daily grind is often hindered by the challenges of knowledge sharing, seeking expertise, and managing technical tasks. In this dynamic environment, efficiency takes a hit as employees grapple with the complexities of day-to-day tasks while juggling multiple software applications. The quest to find solutions to these persistent issues led to the creation of something remarkable - a solution that aimed to revolutionise the way we work and learn.
This case study is a journey into creating a user-friendly AI Work/Learning Assistant. In this case study, we'll go through the hurdles my team found in sharing knowledge at workplaces and how Dew came to the rescue. You'll see how Dew evolved from an idea to a mascot and a solution that seamlessly fits into ones daily tasks, helping them learn and work better, and paving the way for knowledge-led growth.
A little about PaddleBoat…
Expand to read
A little about PaddleBoat…
Expand to read
Problem Statement
Problem Statement
In any organisation, it's often a challenge for junior staff to quickly access senior expertise, leading to time-consuming meetings and delays for technical clarifications. Imagine having an ever-available assistant in the organisation that caters to their specific role, providing instant answers whenever needed. While numerous AI tools exist, finding the right balance between ease of use and role customisation can be complex. AI jargon often makes quick, role-based assistance inaccessible to most employees.
In any organisation, it's often a challenge for junior staff to quickly access senior expertise, leading to time-consuming meetings and delays for technical clarifications. Imagine having an ever-available assistant in the organisation that caters to their specific role, providing instant answers whenever needed. While numerous AI tools exist, finding the right balance between ease of use and role customisation can be complex. AI jargon often makes quick, role-based assistance inaccessible to most employees.
We were able to find 3 main problems with knowledge-sharing & using AI tools:
Challenge in Accessing Expertise
Employees struggle to access senior expertise quickly, leading to time-consuming meetings and delays for technical clarifications.Inaccessibility to assistance due to AI Jargon
AI jargon often renders quick, role-based assistance inaccessible to most employees.Customisations are rare on AI responses
The market is saturated with AI tools, yet finding one that strikes a balance between ease of use and role customisation (job-specific) is rare.
We were able to find 3 main problems with knowledge-sharing & using AI tools:
Challenge in Accessing Expertise
Employees struggle to access senior expertise quickly, leading to time-consuming meetings and delays for technical clarifications.Inaccessibility to assistance due to AI Jargon
AI jargon often renders quick, role-based assistance inaccessible to most employees.Customisations are rare on AI responses
The market is saturated with AI tools, yet finding one that strikes a balance between ease of use and role customisation (job-specific) is rare.
Step 1 - Brain Dump
Step 1 - Brain Dump
Before we built a full blown AI learning assistant, we were curious as to what actually constitutes a good assistant. The team spent a few days to do some quick research around what makes a good assistant. We collected our thoughts, inspirations and points to discuss with the whole team to get a broader understanding on everyone's takes.
After multiple sessions and very helpful insights we came at cross roads on scoping out the first version of the assistant. A few features that stood out from the discussions:
Before we built a full blown AI learning assistant, we were curious as to what actually constitutes a good assistant. The team spent a few days to do some quick research around what makes a good assistant. We collected our thoughts, inspirations and points to discuss with the whole team to get a broader understanding on everyone's takes.
After multiple sessions and very helpful insights we came at cross roads on scoping out the first version of the assistant. A few features that stood out from the discussions:
For Creators:
Assistance for creating content - content generation, tone/delivery manipulations, para-phrasing, source videos, etc.
Smart recommendations and insights from Content Analytics - based on learner activity.
For Creators:
Assistance for creating content - content generation, tone/delivery manipulations, para-phrasing, source videos, etc.
Smart recommendations and insights from Content Analytics - based on learner activity.
For Learners:
Job specific help - persona based responses
Assistance to help track and complete urgent items before deadline - Notification System & Start of Session
Assistance within Learning Experience - quick keyword look-ups, create teachable moments, help with topic understanding
Smart suggestions to improve performance at end of Assessments
For Learners:
Job specific help - persona based responses
Assistance to help track and complete urgent items before deadline - Notification System & Start of Session
Assistance within Learning Experience - quick keyword look-ups, create teachable moments, help with topic understanding
Smart suggestions to improve performance at end of Assessments
For Admins:
Quick Reports and Insights on organisation level activity - scope for improvement on departments, derived metrics from existing analytical data
For Admins:
Quick Reports and Insights on organisation level activity - scope for improvement on departments, derived metrics from existing analytical data
The team really went all out with suggestions, but due to limited technical bandwidths and other product priorities we decided to phase it out and have a very modest set of goals for the first version.
The team really went all out with suggestions, but due to limited technical bandwidths and other product priorities we decided to phase it out and have a very modest set of goals for the first version.
Step 2 - Finalise features
Step 2 - Finalise features
We decided to open up the AI work/learning assistant for all users in the workforce (creators, learners and admins alike). Mainly to understand usage scenarios, style, and understand how often people take help from an AI for work related questions.
We decided to open up the AI work/learning assistant for all users in the workforce (creators, learners and admins alike). Mainly to understand usage scenarios, style, and understand how often people take help from an AI for work related questions.
The features:
One unified chat interface to interact with the AI.
Persona selection - to help customise AI responses to specific job roles.
Quick Shortcuts for each Persona - to allow quick resolutions to usual redundant tasks for specific job roles.
Copy response - copy AI response, code blocks, etc.
Chat History - allowing users to store multiple chats.
Credit count - Show a progress bar to inform users of token usage.
The features:
One unified chat interface to interact with the AI.
Persona selection - to help customise AI responses to specific job roles.
Quick Shortcuts for each Persona - to allow quick resolutions to usual redundant tasks for specific job roles.
Copy response - copy AI response, code blocks, etc.
Chat History - allowing users to store multiple chats.
Credit count - Show a progress bar to inform users of token usage.
Design goals:
Name and Tonality of AI assistant
Mascot/Character design for AI assistant
Theme for AI features to make design scalable for AI functionalities in the future
Simple and AI chat Interface
Design goals:
Name and Tonality of AI assistant
Mascot/Character design for AI assistant
Theme for AI features to make design scalable for AI functionalities in the future
Simple and AI chat Interface
Step 3 - Character Design
Step 3 - Character Design
This was an exciting task for the whole design team where we were tasked to create the face of the suite of AI features that PaddleBoat would offer.
We were curious as to how we should proceed and took inspirations from several Mascots that are renowned to have a quirky vibe and consistent presence across throughout the user experience.
This was an exciting task for the whole design team where we were tasked to create the face of the suite of AI features that PaddleBoat would offer.
We were curious as to how we should proceed and took inspirations from several Mascots that are renowned to have a quirky vibe and consistent presence across throughout the user experience.
Assistant Name
All of PaddleBoat assembled on Miro and contributed to this exercise to find the perfect name, some of us even went ahead and created brand guidelines for the character. After 2 meetings the team decided to call it Dew. YES! it finally had a name - DEW :)
Assistant Name
All of PaddleBoat assembled on Miro and contributed to this exercise to find the perfect name, some of us even went ahead and created brand guidelines for the character. After 2 meetings the team decided to call it Dew. YES! it finally had a name - DEW :)
Giving Dew a face
The design team got together to research and ideate on the appearance, style and limits that Dew might have.
We came up with a mood board to understand mascot design in other industry leaders (Discord, Duolingo, Waze, etc.). This was also a pivotal point for design since we didn't have a particular illustrations style to stick to. So we decided to come up with a solution that worked for both.
Giving Dew a face
The design team got together to research and ideate on the appearance, style and limits that Dew might have.
We came up with a mood board to understand mascot design in other industry leaders (Discord, Duolingo, Waze, etc.). This was also a pivotal point for design since we didn't have a particular illustrations style to stick to. So we decided to come up with a solution that worked for both.
Design language mood board
Design language mood board
After a tons of discussions, scribbles and quarrels, the team was finally making some breakthrough. The decision was to stick to a simple and minimal illustrations style that had gradients in them - giving it a fresh new age product vibe. After a few iterations internally among designers - about deciding on usage of limbs, features, expressions etc., we we're finally nearing the final version for Dew.
After a tons of discussions, scribbles and quarrels, the team was finally making some breakthrough. The decision was to stick to a simple and minimal illustrations style that had gradients in them - giving it a fresh new age product vibe. After a few iterations internally among designers - about deciding on usage of limbs, features, expressions etc., we we're finally nearing the final version for Dew.
And finally…Dew in all it's glory!
And finally…Dew in all it's glory!
Step 4 - The Chat interface
Step 4 - The Chat interface
The final piece to Dew was it's interface. With the feature set sorted, we were ready to build the perfect home for Dew to conjure it's magic.
We started out with researching on a few platforms on what the users are currently used to for interacting with AI. The obvious ones we went through were ChatGPT, Bing Chat, Bard. Some new and tricky implementations we saw were Merlin AI, Mano (both browser extension).
The final piece to Dew was it's interface. With the feature set sorted, we were ready to build the perfect home for Dew to conjure it's magic.
We started out with researching on a few platforms on what the users are currently used to for interacting with AI. The obvious ones we went through were ChatGPT, Bing Chat, Bard. Some new and tricky implementations we saw were Merlin AI, Mano (both browser extension).
First Version - making wireframes and medium-fidelity UI
The team iterated on a few wireframes to decide a probable layout, after a few discussions we finalised a layout and started making the UI for review with the team.
First Version - making wireframes and medium-fidelity UI
The team iterated on a few wireframes to decide a probable layout, after a few discussions we finalised a layout and started making the UI for review with the team.
Dew - First version
Dew - First version
Everyone loved it!… BUT
Few flags were raised by software and product after they went through ChatGPT API pricing model.
Pricing of GPT API
This approach proved to be expensive because of the implementation of New Chats + Chat History. For Dew to provide actual value to the user - it would either mean rapidly depleting Credits (Tokens) for users or hefty Pricing Plans for the client companies.Context Window
GPT maintains a context window for each chat and it keeps sliding to latest chat that the user had with it - this would make the chat experience poor with GPT forgetting context of the chat really quickly.Storing chat history
Since GPT takes previous conversation into consideration to provide contextual responses - chat history would mean PaddleBoat storing all history in their database which is resent to GPT as context - further impacting the token depletion.
Everyone loved it!… BUT
Few flags were raised by software and product after they went through ChatGPT API pricing model.
Pricing of GPT API
This approach proved to be expensive because of the implementation of New Chats + Chat History. For Dew to provide actual value to the user - it would either mean rapidly depleting Credits (Tokens) for users or hefty Pricing Plans for the client companies.Context Window
GPT maintains a context window for each chat and it keeps sliding to latest chat that the user had with it - this would make the chat experience poor with GPT forgetting context of the chat really quickly.Storing chat history
Since GPT takes previous conversation into consideration to provide contextual responses - chat history would mean PaddleBoat storing all history in their database which is resent to GPT as context - further impacting the token depletion.
Step 5 - Recalibrate and Refine
Step 5 - Recalibrate and Refine
We went back to the sketch board to create a win-win for both PaddleBoat and the Client. Following a series of strategic trimmings and careful considerations we came up with a revised feature set. After a green light from product we crafted the final version of Dew with all the considerations in mind.
Presenting…
We went back to the sketch board to create a win-win for both PaddleBoat and the Client. Following a series of strategic trimmings and careful considerations we came up with a revised feature set. After a green light from product we crafted the final version of Dew with all the considerations in mind.
Presenting…
The best AI learning assistant
The best AI learning assistant
Made this sick video as well of course ;)
Made this sick video as well of course ;)
Time to flex them screens
Time to flex them screens
Dew chat interface
1. Credit usage
A counter to show credits utilisation realtime as the user is conversing with Dew. A tooltip to understand how credits consumption mechanism (abstracted by default).
1. Credit usage
A counter to show credits utilisation realtime as the user is conversing with Dew. A tooltip to understand how credits consumption mechanism (abstracted by default).
2. Persona selector
An intuitive persona selector that makes picking roles easy for Dew to play. It's placed in the chat area to make starting persona based chats with Dew a breeze.
2. Persona selector
An intuitive persona selector that makes picking roles easy for Dew to play. It's placed in the chat area to make starting persona based chats with Dew a breeze.
3. Chat controls
A dedicated panel to control all chat level actions including restarting a new chat and also a place were Persona based shortcuts would be found.
3. Chat controls
A dedicated panel to control all chat level actions including restarting a new chat and also a place were Persona based shortcuts would be found.
4. Example Prompts
For the the less AI-savvy folks out there, we decided to give them a starting point. These showed users how to phrase a prompt for both General chat and Persona based chats.
4. Example Prompts
For the the less AI-savvy folks out there, we decided to give them a starting point. These showed users how to phrase a prompt for both General chat and Persona based chats.
Dew in Developer Persona (Short-cuts appear on left)
Dew in Developer Persona (Short-cuts appear on left)
Chat in progress
Chat in progress
Retrospect and Dew's future
Retrospect and Dew's future
The design team conjured Dew's design and gracefully handed it to the dev team, where the magic continued. Unexpected blockers came up during the initial implementation, yet the team quickly rebounded with an alternative solution, ensuring project timelines weren't effected.
We were able to give Dew a face that oozed confidence 😎, and overall it was a fun exercise to work on character design, to set up a design language for an AI feature. More AI features were to be built and in the pipeline, so creating a consistent language for it across the platform was crucial.
The design team conjured Dew's design and gracefully handed it to the dev team, where the magic continued. Unexpected blockers came up during the initial implementation, yet the team quickly rebounded with an alternative solution, ensuring project timelines weren't effected.
We were able to give Dew a face that oozed confidence 😎, and overall it was a fun exercise to work on character design, to set up a design language for an AI feature. More AI features were to be built and in the pipeline, so creating a consistent language for it across the platform was crucial.
What next?
Dew marked the beginning of an exciting journey for PaddleBoat. It was just the opening chapter, leading us to explore how Dew could seamlessly weave in with Paddleboat's existing infrastructure. Remember the first features list? We didn't discard it; we decided to build upon it.
What next?
Dew marked the beginning of an exciting journey for PaddleBoat. It was just the opening chapter, leading us to explore how Dew could seamlessly weave in with Paddleboat's existing infrastructure. Remember the first features list? We didn't discard it; we decided to build upon it.
A gist of the broader vision:
Single click course creation - User would be able to build courses from scratch with just a few phrases, this would allow them to not look at a blank screen every time they try to build something.
Creation assistance - Creators would always have an ally to help them with anything during their creation journey. Help them deliver content in any tone they see fit, allowing them to focus on the content rather than the delivery style, grammar, etc.
New insights from Analytics - Dew would be able to identify patterns in activity that unlock new insights that PaddleBoat doesn't offer off the shelf.
Single click use-case creation - We would expand the phrase to course system to other use-cases that PaddleBoat offers (Assessments, KT Plan, Video Playlists, Wikis, etc.)
more to follow …
All this powered by Dew.
A gist of the broader vision:
Single click course creation - User would be able to build courses from scratch with just a few phrases, this would allow them to not look at a blank screen every time they try to build something.
Creation assistance - Creators would always have an ally to help them with anything during their creation journey. Help them deliver content in any tone they see fit, allowing them to focus on the content rather than the delivery style, grammar, etc.
New insights from Analytics - Dew would be able to identify patterns in activity that unlock new insights that PaddleBoat doesn't offer off the shelf.
Single click use-case creation - We would expand the phrase to course system to other use-cases that PaddleBoat offers (Assessments, KT Plan, Video Playlists, Wikis, etc.)
more to follow …
All this powered by Dew.
Thank you for reading!
Thank you for reading!