Personal AI
Personal AI
Enhanced AI model training usability and optimized chat invitation processes for a more intuitive flow.
Enhanced AI model training usability and optimized chat invitation processes for a more intuitive flow.
Overview
Overview
Overview
Personal AI is a digital extension of anyone’s personal brand. It is an AI model that uses your own data, facts, and opinions to create a responsive and interactive messaging experience that helps people be more productive and deepen relationships. Think of it as your very own ‘Hey, Siri’ that can respond using your own facts, opinions, and style.
Personal AI is a digital extension of anyone’s personal brand. It is an AI model that uses your own data, facts, and opinions to create a responsive and interactive messaging experience that helps people be more productive and deepen relationships. Think of it as your very own ‘Hey, Siri’ that can respond using your own facts, opinions, and style.
Personal AI is a digital extension of anyone’s personal brand. It is an AI model that uses your own data, facts, and opinions to create a responsive and interactive messaging experience that helps people be more productive and deepen relationships. Think of it as your very own ‘Hey, Siri’ that can respond using your own facts, opinions, and style.
Timeline
Timeline
Oct 2023 — Nov 2023
Oct 2023 — Nov 2023
Context
Context
Contract Product Design Role
Contract Product Design Role
My Role
My Role
User Research
UX Design
Visual Design
Engineering Handoff
User Research
UX Design
Visual Design
Engineering Handoff
Team
Team
1 Head of Design
1 Product Designer (me)
1 PM
2 Engineers
1 Head of Design
1 Product Designer (me)
1 PM
2 Engineers
Specific Contributions
Specific Contributions
Specific Contributions
I led the UX/UI design of an updated consumer user journey working on both large and small scale features. I collaborated cross-functionally with the product team, and other stakeholders to drive project’s direction, I conducted user research, developed interactive prototypes, and facilitated seamless engineering handoff.
I led the UX/UI design of an updated consumer user journey working on both large and small scale features. I collaborated cross-functionally with the product team, and other stakeholders to drive project’s direction, I conducted user research, developed interactive prototypes, and facilitated seamless engineering handoff.
I led the UX/UI design of an updated consumer user journey working on both large and small scale features. I collaborated cross-functionally with the product team, and other stakeholders to drive project’s direction, I conducted user research, developed interactive prototypes, and facilitated seamless engineering handoff.
The Problem
The Problem
The Problem
1
1
1
Users face a lot of frustration and friction when fine tuning AI models, leading to a higher user drop off rate during training.
Users face a lot of frustration and friction when fine tuning AI models, leading to a higher user drop off rate during training.
Users face a lot of frustration and friction when fine tuning AI models, leading to a higher user drop off rate during training.
Regardless of how experienced users were with AI concepts, the current training flow was not clear on how long fine tuning a usable model would take or what kind of data is the most useful for this process. With training multiple models at once, if there was any overlap, users would spend a lot of time on redundant steps copying the data over. Overall users felt the AI model training process was too complicated and hard to understand.
Regardless of how experienced users were with AI concepts, the current training flow was not clear on how long fine tuning a usable model would take or what kind of data is the most useful for this process. With training multiple models at once, if there was any overlap, users would spend a lot of time on redundant steps copying the data over. Overall users felt the AI model training process was too complicated and hard to understand.
Regardless of how experienced users were with AI concepts, the current training flow was not clear on how long fine tuning a usable model would take or what kind of data is the most useful for this process. With training multiple models at once, if there was any overlap, users would spend a lot of time on redundant steps copying the data over. Overall users felt the AI model training process was too complicated and hard to understand.
2
2
2
Users have a hard time accessing the platform's core features due to the product's complex UI, resulting in an overall decline in user retention and engagement.
Users have a hard time accessing the platform's core features due to the product's complex UI, resulting in an overall decline in user retention and engagement.
Users have a hard time accessing the platform's core features due to the product's complex UI, resulting in an overall decline in user retention and engagement.
With the app’s rapid changes and introduction of newer features, users have a hard time keeping track of where to access the platform’s core features. There are also multiple access points to compete the same action which while promotes accessibility, confused users as well as took up valuable real estate for other features. The current UI was also not user friendly, and thus caused new users to be overloaded with information during onboarding.
With the app’s rapid changes and introduction of newer features, users have a hard time keeping track of where to access the platform’s core features. There are also multiple access points to compete the same action which while promotes accessibility, confused users as well as took up valuable real estate for other features. The current UI was also not user friendly, and thus caused new users to be overloaded with information during onboarding.
With the app’s rapid changes and introduction of newer features, users have a hard time keeping track of where to access the platform’s core features. There are also multiple access points to compete the same action which while promotes accessibility, confused users as well as took up valuable real estate for other features. The current UI was also not user friendly, and thus caused new users to be overloaded with information during onboarding.
3
3
3
There are a lot of barriers to inviting friends and building a community on the platform, causing user base growth to be stagnant.
There are a lot of barriers to inviting friends and building a community on the platform, causing user base growth to be stagnant.
There are a lot of barriers to inviting friends and building a community on the platform, causing user base growth to be stagnant.
Personal AI wants to create a more user friendly experience for both new and existing users, focusing on a social network model that emphasizes new user acquisition. With this model, users will be able to expand on their personal brand and build an audience for their specific business or personal needs as well as connect with others in various public communities.
Personal AI wants to create a more user friendly experience for both new and existing users, focusing on a social network model that emphasizes new user acquisition. With this model, users will be able to expand on their personal brand and build an audience for their specific business or personal needs as well as connect with others in various public communities.
Personal AI wants to create a more user friendly experience for both new and existing users, focusing on a social network model that emphasizes new user acquisition. With this model, users will be able to expand on their personal brand and build an audience for their specific business or personal needs as well as connect with others in various public communities.
Solution Highlights
Solution Highlights
Solution Highlights
Research
Research
Research
I worked with the product manager to conduct a variety of user research to better understand pain points and see what parts of the platform we needed to prioritize improving first. I sorted through support tickets trying to find common behavior patterns as well as held user interviews to talk with people directly about their experience.
I worked with the product manager to conduct a variety of user research to better understand pain points and see what parts of the platform we needed to prioritize improving first. I sorted through support tickets trying to find common behavior patterns as well as held user interviews to talk with people directly about their experience.
I worked with the product manager to conduct a variety of user research to better understand pain points and see what parts of the platform we needed to prioritize improving first. I sorted through support tickets trying to find common behavior patterns as well as held user interviews to talk with people directly about their experience.
Understanding the common user experience on Personal AI's web platform.
Understanding the common user experience on Personal AI's web platform.
Understanding the common user experience on Personal AI's web platform.
For interviews, some of my questions were:
For interviews, some of my questions were:
For interviews, some of my questions were:
What are specific challenges or frustrations you faced during the AI model training process?
How long does it typically take for you to train a model, and do you find this timeframe reasonable?
How easy or difficult is it for you to find specific functions (e.g., sending a message, starting a new chat, adding a contact)?
What do you think about the placement and visibility of key actions
How easy or difficult was it to understand the value of the app during the onboarding process?
How easy or difficult was it to understand the value of the app during the onboarding process?
How easy or difficult was it to understand the value of the app during the onboarding process?
I wrote out a few user personas to separate user pain points based on use case.
I wrote out a few user personas to separate user pain points based on use case.
I wrote out a few user personas to separate user pain points based on use case.
With user research done, I compared business and user wants to plan how I can balance both pain points.
With user research done, I compared business and user wants to plan how I can balance both pain points.
With user research done, I compared business and user wants to plan how I can balance both pain points.
💼 Business
💼 Business
💼 Business
1
1
1
Restructure AI training and platform onboarding to be more intuitive and user friendly.
Restructure AI training and platform onboarding to be more intuitive and user friendly.
Restructure AI training and platform onboarding to be more intuitive and user friendly.
2
2
2
Promote user journey completion.
Promote user journey completion.
3
3
3
Create a messaging first experience and encourage inviting friends to the platform.
Create a messaging first experience and encourage inviting friends to the platform.
Create a messaging first experience and encourage inviting friends to the platform.
👤 Users
👤 Users
👤 Users
1
1
1
Spend a reasonable amount of time training their models to generate usable responses.
Spend a reasonable amount of time training their models to generate usable responses.
Spend a reasonable amount of time training their models to generate usable responses.
2
2
2
Incorporate AI models more seamlessly into the messaging experience.
Incorporate AI models more seamlessly into the messaging experience.
Incorporate AI models more seamlessly into the messaging experience.
3
3
3
Streamline core feature flows to complete actions more efficiently.
Streamline core feature flows to complete actions more efficiently.
Streamline core feature flows to complete actions more efficiently.
I summarized my research in 3 key findings.
I summarized my research in 3 key findings.
I summarized my research in 3 key findings.
Key Findings:
Key Findings:
Key Findings:
1
1
1
There is no "completion state" for AI model training.
There is no "completion state" for AI model training.
Most users are unsure how long they need to train their models for and how much data they need to input. This lack of knowledge combined with their frustration of getting unusable responses causes them to drop off the training flow.
Most users are unsure how long they need to train their models for and how much data they need to input. This lack of knowledge combined with their frustration of getting unusable responses causes them to drop off the training flow.
2
2
2
Redundant flows can be consolidated for more UI real estate.
Redundant flows can be consolidated for more UI real estate.
Although there are multiple ways to complete an action i.e. inviting friends to chat, only one or two of those flows are mainly used. The team had also been struggling with finding room to add new features, so removing those extra flows will solve two issues at once.
Although there are multiple ways to complete an action i.e. inviting friends to chat, only one or two of those flows are mainly used. The team had also been struggling with finding room to add new features, so removing those extra flows will solve two issues at once.
3
3
3
Every experience on the platform felt separate from each other.
Every experience on the platform felt separate from each other.
Users felt like they were constantly toggling back and forth between different screens to get things done. There should be more of an integration between each feature for a more seamless holistic experience.
Users felt like they were constantly toggling back and forth between different screens to get things done. There should be more of an integration between each feature for a more seamless holistic experience.
With my research findings, I wrote a few "how might we" statements as well as listed out the overall design focuses for these projects. These serve as a high level guide that I can come back to and reference during the design process to check whether or not my solution is answering the problem at hand.
With my research findings, I wrote a few "how might we" statements as well as listed out the overall design focuses for these projects. These serve as a high level guide that I can come back to and reference during the design process to check whether or not my solution is answering the problem at hand.
With my research findings, I wrote a few "how might we" statements as well as listed out the overall design focuses for these projects. These serve as a high level guide that I can come back to and reference during the design process to check whether or not my solution is answering the problem at hand.
💭 How Might We
💭 How Might We
💭 How Might We
How might we design features that work to support the overall AI training process and thus lead to an increase in AI training time per user?
How might we design features that work to support the overall AI training process and thus lead to an increase in AI training time per user?
How might we design features that work to support the overall AI training process and thus lead to an increase in AI training time per user?
How might we simplify the web app’s core features to create a more seamless experience for new and existing users as well as make room for future application updates?
How might we simplify the web app’s core features to create a more seamless experience for new and existing users as well as make room for future application updates?
How might we simplify the web app’s core features to create a more seamless experience for new and existing users as well as make room for future application updates?
How might we create a social media platform like experience that prioritizes user interaction, engagement, and community-building?
How might we create a social media platform like experience that prioritizes user interaction, engagement, and community-building?
How might we create a social media platform like experience that prioritizes user interaction, engagement, and community-building?
🔍 Design Focuses:
🔍 Design Focuses:
🔍 Design Focuses:
1
1
1
Keep Things Simple & Consistent
Keep Things Simple & Consistent
Keep Things Simple & Consistent
2
2
2
Increase Efficiency
Increase Efficiency
Increase Efficiency
3
3
3
Promote Engagement
Promote Engagement
Promote Engagement
4
4
4
Allow for flexibility
Allow for flexibility
Allow for flexibility
Design
Design
Design
Iterating based on user and team feedback
Iterating based on user and team feedback
Iterating based on user and team feedback
Since the team worked in one week sprints, I would start with mid-fidelity wireframes and work my way up to high fidelity. I iterated on my designs based on feedback from the product team as well as from think-aloud usability testing with users. I worked closely with the Head of Design to ensure I was meeting both business and user requirements.
Since the team worked in one week sprints, I would start with mid-fidelity wireframes and work my way up to high fidelity. I iterated on my designs based on feedback from the product team as well as from think-aloud usability testing with users. I worked closely with the Head of Design to ensure I was meeting both business and user requirements.
Since the team worked in one week sprints, I would start with mid-fidelity wireframes and work my way up to high fidelity. I iterated on my designs based on feedback from the product team as well as from think-aloud usability testing with users. I worked closely with the Head of Design to ensure I was meeting both business and user requirements.
After ideating a few solutions and narrowing down on an option with the product team, I created and presented low to mid-fidelity designs to the Head of Design. She helped me look over the finer details and gave me feedback on what to improve. Some of the main critiques I got on my design process was to double check that everything fit the updated branding guidelines and to watch for consistency when updating older features to match the new UI. An overall key focus was to make sure the designs were intuitive and easy to use for all users.
After ideating a few solutions and narrowing down on an option with the product team, I created and presented low to mid-fidelity designs to the Head of Design. She helped me look over the finer details and gave me feedback on what to improve. Some of the main critiques I got on my design process was to double check that everything fit the updated branding guidelines and to watch for consistency when updating older features to match the new UI. An overall key focus was to make sure the designs were intuitive and easy to use for all users.
After ideating a few solutions and narrowing down on an option with the product team, I created and presented low to mid-fidelity designs to the Head of Design. She helped me look over the finer details and gave me feedback on what to improve. Some of the main critiques I got on my design process was to double check that everything fit the updated branding guidelines and to watch for consistency when updating older features to match the new UI. An overall key focus was to make sure the designs were intuitive and easy to use for all users.
With a need to work and improve the platform quickly in an agile process, I balanced working on large and small scale features depending on what would be the most effective change in the shortest timeframe, and what the other product team members were working on. The chosen solutions for each problem included:
With a need to work and improve the platform quickly in an agile process, I balanced working on large and small scale features depending on what would be the most effective change in the shortest timeframe, and what the other product team members were working on. The chosen solutions for each problem included:
With a need to work and improve the platform quickly in an agile process, I balanced working on large and small scale features depending on what would be the most effective change in the shortest timeframe, and what the other product team members were working on. The chosen solutions for each problem included:
1.1
1.1
1.1
Incorporating an accuracy threshold slider into an AI model's menu
Incorporating an accuracy threshold slider into an AI model's menu
Incorporating an accuracy threshold slider into an AI model's menu
This feature shows the limitations a user’s AI persona currently has. For example, if a user puts in a small amount of training, the slider will only show a 0–25% accuracy range, indicating more training is needed if the user wishes to increase the response accuracy. As the user keeps training, the range will also increase.
This feature shows the limitations a user’s AI persona currently has. For example, if a user puts in a small amount of training, the slider will only show a 0–25% accuracy range, indicating more training is needed if the user wishes to increase the response accuracy. As the user keeps training, the range will also increase.
This feature shows the limitations a user’s AI persona currently has. For example, if a user puts in a small amount of training, the slider will only show a 0–25% accuracy range, indicating more training is needed if the user wishes to increase the response accuracy. As the user keeps training, the range will also increase.
1.2
1.2
1.2
Assigning documents to multiple AI personas
Assigning documents to multiple AI personas
Assigning documents to multiple AI personas
This feature allows users to select all the personas they wish to upload a piece of data to in one go. Users can go in and adjust this setting if they want to add or remove a persona later on. Users can cut down their AI training time and effort if they are building multiple personas on the platform.
This feature allows users to select all the personas they wish to upload a piece of data to in one go. Users can go in and adjust this setting if they want to add or remove a persona later on. Users can cut down their AI training time and effort if they are building multiple personas on the platform.
This feature allows users to select all the personas they wish to upload a piece of data to in one go. Users can go in and adjust this setting if they want to add or remove a persona later on. Users can cut down their AI training time and effort if they are building multiple personas on the platform.
2
2
2
Combining chatting and invite flows
Combining chatting and invite flows
Combining chatting and invite flows
I listed all the primary actions in one menu that users can navigate to. Each option opens a pop up window that guides users through the next actions to take, and I reduced the steps to be as few as possible so that users reach their goal a lot quicker. I also updated the invite feature to account for various types of users as well as incorporated an invite link that can be copied and pasted.
I listed all the primary actions in one menu that users can navigate to. Each option opens a pop up window that guides users through the next actions to take, and I reduced the steps to be as few as possible so that users reach their goal a lot quicker. I also updated the invite feature to account for various types of users as well as incorporated an invite link that can be copied and pasted.
I listed all the primary actions in one menu that users can navigate to. Each option opens a pop up window that guides users through the next actions to take, and I reduced the steps to be as few as possible so that users reach their goal a lot quicker. I also updated the invite feature to account for various types of users as well as incorporated an invite link that can be copied and pasted.
3
3
3
Centralized chatting
Centralized chatting
Centralized chatting
Given that most users chat with a small number of people/are in a few lounges, I combined DMs and lounges so that users can access both on one screen. Users can also pin their most active conversations to the top of their messaging list for easier access. I also included the discover feature on this side menu for users to easily access and find other people/lounges.
Given that most users chat with a small number of people/are in a few lounges, I combined DMs and lounges so that users can access both on one screen. Users can also pin their most active conversations to the top of their messaging list for easier access. I also included the discover feature on this side menu for users to easily access and find other people/lounges.
Given that most users chat with a small number of people/are in a few lounges, I combined DMs and lounges so that users can access both on one screen. Users can also pin their most active conversations to the top of their messaging list for easier access. I also included the discover feature on this side menu for users to easily access and find other people/lounges.
Once the designs were finalized, I created a file containing the high fidelity mockups to be sent to the development team. I walked through the designs in a Loom recording for the development team to reference, breaking down the solution and explaining any new or updated features.
Once the designs were finalized, I created a file containing the high fidelity mockups to be sent to the development team. I walked through the designs in a Loom recording for the development team to reference, breaking down the solution and explaining any new or updated features.
Once the designs were finalized, I created a file containing the high fidelity mockups to be sent to the development team. I walked through the designs in a Loom recording for the development team to reference, breaking down the solution and explaining any new or updated features.
Conclusion
Conclusion
Conclusion
Impact
Impact
Impact
AI model training time decreased by over 80% for users
AI model training time decreased by over 80% for users
The data upload feature alongside an improved AI model training process was able to support users' training efforts by decreasing the amount of time and effort it would take to both create multiple models and generate usable responses for each of them.
The data upload feature alongside an improved AI model training process was able to support users' training efforts by decreasing the amount of time and effort it would take to both create multiple models and generate usable responses for each of them.
Reduced task duration by 20% across the whole platform
Reduced task duration by 20% across the whole platform
By eliminating redundant flows, consolidating various features, and simplifying the platform's overall user interface, users were able to complete tasks with less friction and in fewer steps.
By eliminating redundant flows, consolidating various features, and simplifying the platform's overall user interface, users were able to complete tasks with less friction and in fewer steps.
Eliminated user confusion on AI training
Eliminated user confusion on AI training
The accuracy threshold slider in conjunction with an improved, user friendly onboarding process was able to better communicate AI training concepts to users. Once the feature shipped, there were little to no support tickets being sent in by users on clarifying limitations and capabilities in regards to their model.
The accuracy threshold slider in conjunction with an improved, user friendly onboarding process was able to better communicate AI training concepts to users. Once the feature shipped, there were little to no support tickets being sent in by users on clarifying limitations and capabilities in regards to their model.
Next Steps
Next Steps
Next Steps
Longer term user research and usability testing
Longer term user research and usability testing
Although the developed solutions improved the platform's overall user experience, further testing should be done to ensure these are the best solutions for users, especially for the features surrounding AI model training. With more data, I would want to continue iterating based on user feedback.
Although the developed solutions improved the platform's overall user experience, further testing should be done to ensure these are the best solutions for users, especially for the features surrounding AI model training. With more data, I would want to continue iterating based on user feedback.
Monitor and improve by looking at KPIs (drop off rates, user acquisition rate, model training completion rate, etc.)
Monitor and improve by looking at KPIs (drop off rates, user acquisition rate, model training completion rate, etc.)
There were a lot of metrics the team wanted to improve upon when I first joined Personal AI. A few of the projects I worked on were part of newer initiatives i.e. creating a messaging first experience, so it would be beneficial to see how certain KPIs were affected once the whole initiative was completed, not just a few features.
There were a lot of metrics the team wanted to improve upon when I first joined Personal AI. A few of the projects I worked on were part of newer initiatives i.e. creating a messaging first experience, so it would be beneficial to see how certain KPIs were affected once the whole initiative was completed, not just a few features.
Targeting more specific user groups
Targeting more specific user groups
Simplifying the platform's user interface allowed Personal AI's core features to shine, but the downside is that the user experience has become more generalized. Personal AI has a diverse user base where users are creating AI models for various specific needs. I would want to start exploring how to best support these use cases now that the platform's core features haven been strengthened.
Simplifying the platform's user interface allowed Personal AI's core features to shine, but the downside is that the user experience has become more generalized. Personal AI has a diverse user base where users are creating AI models for various specific needs. I would want to start exploring how to best support these use cases now that the platform's core features haven been strengthened.
Connect with me!
I love making new friends and getting to know people. Feel free to reach out to me if you have any questions or just want to chat!
Made with love and care by Crystal Huynh © 2024
Connect with me!
I love making new friends and getting to know people. Feel free to reach out to me if you have any questions or just want to chat!
Made with love and care by Crystal Huynh © 2024
Connect with me!
I love making new friends and getting to know people. Feel free to reach out to me if you have any questions or just want to chat!
Made with love and care by Crystal Huynh © 2024
Connect with me!
I love making new friends and getting to know people. Feel free to reach out to me if you have any questions or just want to chat!
Made with love and care by Crystal Huynh © 2024