Building empathy into AI: How we designed a Voiceflow chatbot for autism support
A calm, trustworthy chatbot that helps parents of newly diagnosed children understand their options, without judgment and without overwhelm.
The problem
Parents feel overwhelmed and alone while searching for answers
When a child is first diagnosed with autism, parents often find themselves suddenly thrust into a world filled with jargon, fragmented resources, and impossible decisions. Between late-night Googling, closed Facebook groups, and trying to make sense of the special education system, most parents end up asking the same thing:
“Why can’t someone just tell me what to do next?”
Background & role
Femmecubator is a Civic Tech and data driven non-profit based in NYC. I was onboarded as a Sr. UX Designer to assist their AI Chatbot initiative in collaboration with UC San Diego’s Center for Autism. This project aims to serve 240k+ children in Southern California, and expand to 2M+ affected families across the United States.
- Led UX research (interviews, empathy mapping, persona synthesis)
- Created landing page IA and wireframes
- Mapped chatbot architecture, persona and created chatbot prototype in Voiceflow, integrating NLU intents and flows
What we set out to design
A chatbot you could just text your scenario to, and it would help you understand your options. Without judgment, without overwhelm.
They don’t need another 50 page guide to read through. The bot replies with information relevant to their state, their child’s age, and their emotional state. They need a calm, trustworthy voice that listens and gives them answers tailored to their situation.
DEFINING AI AGENTS
The agent is designed to guide parents with clear, supportive paths. It uses simple rules and smart intent jumps to keep conversations easy and helpful.
INTENT MAPPING
Each intent is clearly defined to cover different types of support parents may need. Strong utterance matches help the agent understand questions easily and keep conversations on track.
Key design decisions
Over six weeks, I led the UX research and design of this AI-driven chatbot and its companion landing page.
Every decision, from tone to layout was rooted in real stories from parents.
We used research insights to guide the content structure, flow logic, and even how the bot responds when someone’s feeling overwhelmed.
1. NLU for responses, LLMs for sentiment
We built the chatbot using structured Natural Language Understanding (NLU) and used Large Language Models (LLM) to generate introduction texts, match voice & tone for user queries.
2. Emotionally adaptive
The chatbot was designed to recognize emotional cues such as stress, confusion, or urgency and adjust its tone accordingly, thanks to LLMs!
3. Building an extensive knowledge base
Instead of offering generic advice, our chatbot delivered precise, relevant information drawn from sources verified by our team at UC San Diego to help parents find answers that truly matched their child’s needs and circumstances.
The result
A working prototype built in Voiceflow that feels more like texting a helpful friend. Focused on empathy, built with NLU, and designed to meet parents where they are.
Next steps
Test the prototype with more parent users, launch the website and roll out a beta version, and integrate Twilio APIs for SMS, WhatsApp and Messenger.
Personal reflections
It isn’t easy to build AI chatbots that deal with complex emotions
60% of the project was defining content blocks since we constantly kept missing utterances, or failed to map the right emotions and user sentiments. I learned how crucial tone, transparency, and giving users control are in helping people navigate stress and uncertainty.
Reach out to your peers! This project pushed me to connect with Conversation Designers across my network to help me troubleshoot any issues I had, or simply to learn best practices and processes. I now have a large community to lean on and build with that I am very grateful for!
