Speak Naturally
Users should not need to memorize rigid commands or adapt to machine-like syntax.
Assistivity
Assistivity
ROLE
Co-Founder & Design Lead
TIMELINE
January 2025 - Present
TEAM
1 PM,
1 Engineer,
1 Designer (me!)
SKILLS
Product Design, Development, User Research, Collaboration
ROLE
Co-Founder & Design Lead
TIMELINE
January 2025 - Present
TEAM
1 PM,
1 Engineer,
1 Designer (me!)
SKILLS
Product Design, Development, User Research, Collaboration
ROLE
Co-Founder & Design Lead
TIMELINE
January 2025 - Present
TEAM
1 PM,
1 Engineer,
1 Designer (me!)
SKILLS
Product Design, Development, User Research, Collaboration
OVERVIEW
Creating an AI Voice to Math workspace to help students write math naturally.
We are designing and building an AI-driven workspace that lets users dictate math naturally and explore relationships visually. The project blends research, design, and voice interfaces to make math more accessible and focused on understanding, not mechanics.
Creating an AI Voice to Math workspace to help students write math naturally.
We are designing and building an AI-driven workspace that lets users dictate math naturally and explore relationships visually. The project blends research, design, and voice interfaces to make math more accessible and focused on understanding, not mechanics.
Creating an AI Voice to Math workspace to help students write math naturally.
We are designing and building an AI-driven workspace that lets users dictate math naturally and explore relationships visually. The project blends research, design, and voice interfaces to make math more accessible and focused on understanding, not mechanics.
Note: This is a condensed snapshot of the project rather than a full case study, some process and details have been intentionally left out.
OPPORTUNITY
When Thinking Is Faster Than Typing.
Typing math is frustrating and presents real challenges for students with fine motor disabilities. But it's not only users with accessibility needs, writing precise mathematics is difficult for everyone.
When Thinking Is Faster Than Typing.
Typing math is frustrating and presents real challenges for students with fine motor disabilities. But it's not only users with accessibility needs, writing precise mathematics is difficult for everyone.
When Thinking Is Faster Than Typing.
Typing math is frustrating and presents real challenges for students with fine motor disabilities. But it's not only users with accessibility needs, writing precise mathematics is difficult for everyone.
We saw an opportunity to make math creation as natural as speaking.
WHY THIS MATTERS
Math Should be Accessible to Everyone
This project began with a personal and immediate problem. Our co-founder was highly capable in mathematics, but a wrist injury made it painful and difficult to physically write equations. Despite having the knowledge and ability to succeed, the act of writing became a barrier.
Math Should be Accessible to Everyone
This project began with a personal and immediate problem. Our co-founder was highly capable in mathematics, but a wrist injury made it painful and difficult to physically write equations. Despite having the knowledge and ability to succeed, the act of writing became a barrier.
Math Should be Accessible to Everyone
This project began with a personal and immediate problem. Our co-founder was highly capable in mathematics, but a wrist injury made it painful and difficult to physically write equations. Despite having the knowledge and ability to succeed, the act of writing became a barrier.
That experience became a strong motivation for us:
As we explored the broader opportunity, it became clear that many others could benefit from a hands-free math tool, including students, educators, researchers, and people navigating temporary or long-term physical limitations.
SOLUTION
Turning Spoken Math Into Structured Equations
We designed a hands-free math tool that allows users to create and edit mathematical expressions using voice. Instead of relying on keyboards, equation editors, or handwriting, users can speak naturally while the system converts their input into properly formatted math in real time.
Turning Spoken Math Into Structured Equations
We designed a hands-free math tool that allows users to create and edit mathematical expressions using voice. Instead of relying on keyboards, equation editors, or handwriting, users can speak naturally while the system converts their input into properly formatted math in real time.
Turning Spoken Math Into Structured Equations
We designed a hands-free math tool that allows users to create and edit mathematical expressions using voice. Instead of relying on keyboards, equation editors, or handwriting, users can speak naturally while the system converts their input into properly formatted math in real time.
Early Traction & Accomplishments
Although still in an early stage, the project has already generated strong momentum across research, funding, and institutional interest.
Early Traction & Accomplishments
Although still in an early stage, the project has already generated strong momentum across research, funding, and institutional interest.
Early Traction & Accomplishments
Although still in an early stage, the project has already generated strong momentum across research, funding, and institutional interest.
From accessibility research recognition to university pilot commitments, these milestones helped validate both the problem space and the potential impact of the product.
STARTING FROM ZERO
Finding the Right Questions to Ask
When we began, we didn't have an established interaction model for voice-driven mathematical input. Before designing interfaces, we first needed to understand how users naturally speak math, where existing tools created friction, and what a hands-free workflow should actually feel like.
Finding the Right Questions to Ask
When we began, we didn't have an established interaction model for voice-driven mathematical input. Before designing interfaces, we first needed to understand how users naturally speak math, where existing tools created friction, and what a hands-free workflow should actually feel like.
Finding the Right Questions to Ask
When we began, we didn't have an established interaction model for voice-driven mathematical input. Before designing interfaces, we first needed to understand how users naturally speak math, where existing tools created friction, and what a hands-free workflow should actually feel like.
y = 2x + 3
“two x plus three equals y”
“y equals two x plus three”
“y is proportional to x with offset three”
“a line with slope 2 and intercept 3”
RESEARCH
From Exploration to Published Research
As our understanding of the problem space evolved, we wanted a structured way to synthesize our findings and evaluate existing approaches. We formalized this work through a research paper that explored opportunities for voice-driven mathematical input and reviewed the limitations of current solutions. The paper was later accepted and presented at ACM ASSETS 2025.
From Exploration to Published Research
As our understanding of the problem space evolved, we wanted a structured way to synthesize our findings and evaluate existing approaches. We formalized this work through a research paper that explored opportunities for voice-driven mathematical input and reviewed the limitations of current solutions. The paper was later accepted and presented at ACM ASSETS 2025.
From Exploration to Published Research
As our understanding of the problem space evolved, we wanted a structured way to synthesize our findings and evaluate existing approaches. We formalized this work through a research paper that explored opportunities for voice-driven mathematical input and reviewed the limitations of current solutions. The paper was later accepted and presented at ACM ASSETS 2025.
PRODUCT PRINCIPLES
Turning Ambiguity into Product Direction
We explored many possibilities, but over time certain principles consistently emerged and guided our decisions.
Turning Ambiguity into Product Direction
We explored many possibilities, but over time certain principles consistently emerged and guided our decisions.
Turning Ambiguity into Product Direction
We explored many possibilities, but over time certain principles consistently emerged and guided our decisions.
Keep Users In Their Flow
Math input should support thinking speed instead of interrupting it.
Make Feedback Immediate
Real-time rendering helps users build trust and confidence while speaking.
Reduce Correction Friction
Editing equations should feel lightweight and recoverable rather than punishing.
Prioritize Accessibility By Default
The experience should remain usable for people with different physical abilities and interaction needs.
Because the product space was still evolving, many parts of the experience changed rapidly throughout development. Rather than anchoring too early on fixed interfaces, we focused on identifying the core principles that consistently made the experience feel more natural, accessible, and intuitive.
Through iterative exploration, conversations, and testing, these principles became the foundation for how we approached the product.
CURRENT STATE
Where We Are Today
Assistivity is now live and in the hands of real users, with onboarding and core workspace experiences shipped.
Where We Are Today
Assistivity is now live and in the hands of real users, with onboarding and core workspace experiences shipped.
Where We Are Today
Assistivity is now live and in the hands of real users, with onboarding and core workspace experiences shipped.
System in Motion
Motion isn't just for the hero moments. Across the app bar, sidebar, and everyday actions like saving and exporting, small animations help the product feel responsive, clear, and in flow with the user.
System in Motion
Motion isn't just for the hero moments. Across the app bar, sidebar, and everyday actions like saving and exporting, small animations help the product feel responsive, clear, and in flow with the user.
System in Motion
Motion isn't just for the hero moments. Across the app bar, sidebar, and everyday actions like saving and exporting, small animations help the product feel responsive, clear, and in flow with the user.
THE MATH NODE
Where Voice Becomes Math
The math node is the core interaction of Assistivity, where spoken input becomes structured math in real time. Below is a closer look at how it works and the details that shape the experience.
Where Voice Becomes Math
The math node is the core interaction of Assistivity, where spoken input becomes structured math in real time. Below is a closer look at how it works and the details that shape the experience.
Where Voice Becomes Math
The math node is the core interaction of Assistivity, where spoken input becomes structured math in real time. Below is a closer look at how it works and the details that shape the experience.
Anatomy of a Node
Each node is made up of two parts: a top bar for quick actions like copying, recoloring, and deleting, and the node body itself, where users press the microphone to start speaking their equation.
Anatomy of a Node
Each node is made up of two parts: a top bar for quick actions like copying, recoloring, and deleting, and the node body itself, where users press the microphone to start speaking their equation.
From Voice to Text
As you speak, your equation appears as plain text inside the node, giving you a chance to review what was heard before committing to it.
From Voice to Text
As you speak, your equation appears as plain text inside the node, giving you a chance to review what was heard before committing to it.
From Text to Math
Once submitted, the plain text transforms into a properly formatted equation, structured, readable, and ready to build on.
From Text to Math
Once submitted, the plain text transforms into a properly formatted equation, structured, readable, and ready to build on.
EQUATION CONTROLS
View the LaTeX
Every rendered equation can be flipped to its underlying LaTeX, giving users a way to inspect, copy, or verify the exact structure behind what they see.
View the LaTeX
Every rendered equation can be flipped to its underlying LaTeX, giving users a way to inspect, copy, or verify the exact structure behind what they see.
View the LaTeX
Every rendered equation can be flipped to its underlying LaTeX, giving users a way to inspect, copy, or verify the exact structure behind what they see.
Edit by voice
Equations can be edited the same way they were created, by speaking. This keeps corrections lightweight and in flow, without forcing users into a keyboard or equation editor.
Edit by voice
Equations can be edited the same way they were created, by speaking. This keeps corrections lightweight and in flow, without forcing users into a keyboard or equation editor.
Edit by voice
Equations can be edited the same way they were created, by speaking. This keeps corrections lightweight and in flow, without forcing users into a keyboard or equation editor.
Add the next equation
A quick control lets users add a new equation directly below the current one, treating it as the next step. It keeps the flow of thinking uninterrupted between related equations.
Add the next equation
A quick control lets users add a new equation directly below the current one, treating it as the next step. It keeps the flow of thinking uninterrupted between related equations.
Add the next equation
A quick control lets users add a new equation directly below the current one, treating it as the next step. It keeps the flow of thinking uninterrupted between related equations.
Handling ambiguity
Spoken math isn't always precise. When an equation could be interpreted more than one way, the node highlights it and offers the possible options, letting users pick the one they meant.
Handling ambiguity
Spoken math isn't always precise. When an equation could be interpreted more than one way, the node highlights it and offers the possible options, letting users pick the one they meant.
Handling ambiguity
Spoken math isn't always precise. When an equation could be interpreted more than one way, the node highlights it and offers the possible options, letting users pick the one they meant.
Error states
When something can't be rendered, the node shifts into an error state that surfaces what went wrong, so users can recover quickly without guessing.
Error states
When something can't be rendered, the node shifts into an error state that surfaces what went wrong, so users can recover quickly without guessing.
Error states
When something can't be rendered, the node shifts into an error state that surfaces what went wrong, so users can recover quickly without guessing.
REFLECTION
What I have learned so far
Accessibility as a lens
This project shifted how I think about accessibility in design. Rather than treating it as a constraint layered onto a finished experience, I learned to treat it as a starting point, one that often leads to better, more intentional design for everyone.
Research shaping design
This project pushed me to move beyond designing from intuition and into designing from research. Formalizing our thinking through a published paper reframed how I approach ambiguity, from something to solve quickly, to something worth studying carefully before shaping a product around it.
Accessibility as a lens
This project shifted how I think about accessibility in design. Rather than treating it as a constraint layered onto a finished experience, I learned to treat it as a starting point, one that often leads to better, more intentional design for everyone.
Research shaping design
This project pushed me to move beyond designing from intuition and into designing from research. Formalizing our thinking through a published paper reframed how I approach ambiguity, from something to solve quickly, to something worth studying carefully before shaping a product around it.
Designing a new interaction model
Designing for voice-driven math meant working without an established playbook. It taught me to lean on principles instead of patterns, and to trust that a strong interaction model can emerge from ambiguity when the thinking behind it is clear.
Designing a new interaction model
Designing for voice-driven math meant working without an established playbook. It taught me to lean on principles instead of patterns, and to trust that a strong interaction model can emerge from ambiguity when the thinking behind it is clear.