< Previous ProjectNext Project >
AI AAC Device for Children with Autism
Imagining a schedule-first AI AAC prototype for children with autism and their caregivers
Context
As part of University of Washington's research-heavy academic project, my team and I explored how AI could improve Augmentative and Alternative Communication (AAC) for autistic children.
Team
Jayden Kang
Henson Chen
Kathryn Rambo
Timeline
July - September 2025
Ongoing Project
Key Contributions
Research Synthesis
Turned survey responses into actionable design insights.
Survey & Flow Design
Structured caregiver survey for clear, usable results
Framing & Ideation Leadership
Led research question synthesis and “How Might We” sessions.
Prototype & Testing Design
Built schedule-first prototype and usability testing script.
SUmmary to save you your time
tl;dr
Here’s a quick TL;DR of our research and design process. Each card links to the full section.
Problem
Children with autism struggle to express abstract needs; caregivers are left guessing, and existing AAC tools don’t adapt to time, routines, or emotions.
Research
We surveyed and interviewed 19 caregivers. A clear pattern emerged: children’s ability to communicate changes depending on time of day and daily routines, yet current AAC devices don’t adapt to these shifts.
Insight
Mornings and transitions were consistently reported as the hardest moments. Caregivers said that when routines broke down, communication success dropped sharply. Families needed tools that could scaffold daily structure and anticipate these challenges.
Solution
We designed a schedule-first, AI-powered AAC prototype that integrates Now/Next routines, adaptive phrase suggestions, emotion recognition, and calming tools. It aims to reduce caregiver guesswork and help children express needs before escalation.
Next Steps
This is an ongoing project. Currently, we are done with the pilot testing of the survey and the low fidelity prototype. We will be conducting user testing and more qualitative user interviews in our next steps.
“I hope technology can help my child express abstract feelings and sensations—like hunger, tiredness, or poor sleep."
‍
- Caregiver Response from Survey
Context & Problem
What are AAC devices?
AAC devices are tools that support communication when speech is difficult — ranging from picture cards to speech-generating apps.

‍Current AAC devices are static, hard to personalize, and often abandoned.
Research Approach
Grounded in Literature, Informed by Caregivers
Affinity Mapping
We began with a literature review of AAC. This helped us map where current tools succeed — like providing alternatives to speech — and where they fail: rigid boards, low adoption, and high caregiver effort.

We then affinity-mapped insights from the review into broad themes:Device limitations, Emotional context, Caregiver needs, Opportunities for AI integration
Survey
These themes informed the design of our caregiver survey and questionnaire. We reached 19 parents of children with autism (ages 5–25), using a mix of Likert-scale ratings and open-ended qualitative questions to capture both measurable patterns and personal stories about daily communication.
analyzing data
Key Findings
Core Caregiver Frustrations
01 Children’s rigid routines
“He insists on finding specific brands (e.g., Panasonic) when shopping and is obsessively repetitive.” – Mother of a 5-year-old boy with ASD
02 Difficulty with abstract concepts and schedules
“My child repeatedly asks about the day’s schedule until I write it down!” – Mother of a 25-year-old man with ASD
03 Meltdowns tied to time of day
“He gets angry and throws things when he can't open a bottle—especially right after waking up or when overtired.” – Mother of a 5-year-old boy with ASD
Other Findings
Low adoption of AAC
Only ~20% of participants are using AAC currently
- Concern that AAC might inhibit some child's natural verbal speech development
- Complexity, cost, and lack of professional guidance
- limited adaptability and overwhelming for kids
Current successful strategies
- Visual schedules
- Simplify the language
- Leverage the child's special interests (e.g. cartoon characters)
Question B12 - Using drawings or visual content(e.g. cards with pictures on them, communication boards)greatly improves my ability to communicate with my child.
Together, these findings led us to a clear problem:
Children with autism struggle to express abstract needs; caregivers are left guessing, and existing AAC tools don’t adapt to time, routines, or emotions.
Thinking about next steps
Hypothesis of Interest
Preparing for our next round of user research
Based on our first caregiver study, we formulated hypotheses to guide our next survey. We wanted to test how contextual factors — such as caregiver experience and fit of current tools — shape openness to change and trust in AI-powered AAC.
Trust in AI
If caregivers feel their current communication method works well -> they are more concerned that AI will make AAC too complex (rho=0.68)‍
If caregivers feel the current communication method doesn’t fit their child’s needs -> they are more comfortable sharing video/audio data with AI for model training (rho=-0.57)‍
If children struggle to initiate conversations -> their caregivers are more likely to trust AI to support communication (rho=0.61)
Openness to change
If caregivers feel the current communication method doesn’t fit their child’s needs -> they are more open to change on communication methods (rho=-0.74)‍
If caregivers are older or have longer caregiver experience -> they are less open to change (rho=-0.72)‍
These hypotheses will shape our next caregiver survey, where we aim to validate how caregiver context influences adoption and trust in AI-powered AAC.
Exploring the Solution Space
What if AAC could anticipate needs, support routines, and respond to emotional context, rather than staying static?
Team Ideation Session
After synthesizing insights and forming our hypothesis, we reframed the problem into opportunities.

I led a team ideation session where we generated and clustered “How Might We” questions, ensuring they tied directly back to caregiver needs and our research themes.
HMW help children express internal needs before escalation?

HMW we adapt AAC to daily rhythms and emotions?

HMW reduce caregiver guesswork while maintaining trust?
The Turning Point: A Schedule-First, AI-Powered Prototype
Our research showed time of day strongly shaped communication — mornings were harder, transitions caused stress, and routines gave children stability.

To explore this, we built a speculative prototype in v0.dev — not as a final app, but as a research probe to test whether a schedule-centered AAC approach could reduce caregiver guesswork and better support families.
01 Now/Next Routines + Predictive Assistance
AI surfaces phrases at the right time (e.g., “I’m hungry” at mealtime) → children express needs earlier.
02 Daily Emotion Check - Emotion Recognition + Correction Loop
AI labels emotions, caregivers refine → builds emotional vocabulary and reduces guesswork.
03 Emergency Mode
One-tap shortcut to calming tools → de-escalates meltdowns instantly.
04 Caregiver Dashboard
Tracks progress + offers AI insights → caregivers gain clarity and control.
AI thinking layer
AI differentiates this prototype from traditional AAC.
What AI enables:
Predictive suggestions → anticipates likely needs by time/routine.
Context adaptation → surfaces phrases tied to environment and schedule.
Emotion loop → adaptive learning through caregiver corrections.
Risks flagged:
Mislabeling → caregiver trust issues.
Overreliance → risk of tech replacing human bonding.
Privacy → sensitive comms patterns require protection.
What are we up to?
Next Steps
We plan to move beyond speculation into testing:
Looking back...
Reflection
Designing with/for AI
Learned to balance assistive potential with caregiver trust.
Framed AI as supportive partner, not replacement.
Designing with/for AI
Learned to balance assistive potential with caregiver trust.
Framed AI as supportive partner, not replacement.
Challenges
Short project timeframe
Limited knowledge on subject area
Some survey questions caused confusion
What we would do differently
Allow more time for literature review to strengthen survey design
Extend preparation period to refine questions and methods before data collection
Conduct 2-3 preliminary test surveys before the full pilot.