01 — Project details
Role: Conversation Designer
Platform: Internal healthcare app chatbot
Goal: Create a conversational assistant to support pregnant patients between appointments by helping them get trusted information, triage symptoms, and ease anxiety through human-centered dialogue.
02 — The Challenge
The client wanted a way to help patients anxiety in between checkups due to problems like unease about symptoms and uncertainty on when to contact their provider.
The team and I workshopped with them to design a chatbot that could bridge that gap: offering reassurance, triage guidance, and empathetic conversation while maintaining clinical accuracy and accessibility.
03 — My Role
I led the end-to-end conversation design process:
Researched common patient questions, emotional needs during pregnancy, and credible sources.
Created intent map, conversation goals, and safeguards aligned with healthcare triage standards.
Engineered prompt with extensive tone and personality guidelines emphasizing warmth, clarity, and inclusivity.
Designed and scripted the dialogue flow, including branching logic and edge cases.
Developed testing and QA scoring for refinement.
Initial Prompt
After research & intent mapping
04 — Designing Conversation Flow
After research standard pregnancy triage framework, I designed branching flows to address:
Common symptoms (fatigue, nausea, mild cramps)
Concerning symptoms (bleeding, severe pain, decreased movement)
Emotional check-ins (“I’m feeling anxious,” “I can’t sleep,” etc.)
Each path ended with clear next steps for normal settings (calling a provider, at-home monitoring, accessing educational content) or fallback scenarios if the assistant is ever unclear about user intent or the correct response (“I want to make sure you get the right help. Since I’m not sure how to answer this accurately, it may be best to check in with your provider.”)
To ensure user safety, I included escalation triggers for urgent cases and designed exit conditions that smoothly transitioned users from the chatbot to human support when necessary.
05 — Escalation Triggers
If the user reports any of the below, the assistant must inform the user to contact their prenatal provider immediately, go to the ER, or call 911.
Vaginal bleeding or passing tissue; heavy bleeding.
Sudden or severe abdominal/pelvic pain.
Decreased or absent fetal movement (after ~20 weeks) or marked drop from baseline.
Severe or persistent headache, blurred vision, or sudden swelling of face/hands (possible preeclampsia).
Fever ≥100.4°F (38°C) or fever with pain/chills.
Chest pain, trouble breathing, fainting, severe dizziness, or signs of a blood clot (leg pain/swelling).
Thoughts of hurting yourself or your baby — urgent mental-health escalation.
These trigger examples are drawn from authoritative guidance: ACOG, CDC/AIM, Mayo Clinic, NHS, and WHO/NCBI summaries.
06 — Exit Conditions
Emergency Escalation
If a user describes severe or concerning symptoms, stop all educational dialogue and respond with an urgent care recommendation.
Provider Escalation
If the question involves specific medical advice or diagnosis, redirect to the patient’s healthcare provider.
Conversation Closure
End politely when the user expresses gratitude, satisfaction, or no further questions.
Fallback
If misunderstanding persists after 2 clarification attempts, apologize and suggest professional follow-up.
Ethical Boundaries
Avoid topics like abortion, gender prediction, or medication dosing and redirect to licensed professionals.
07 — Scenario Testing
08 — QA Scoring Rubric (1-5 rating with A/B prompt versioning)
Tone & Voice
Uses a calm, empathetic tone (e.g., “That’s a really important question.”)
Uses plain, accessible language; no jargon or abbreviations
Validates emotions or effort (“It sounds like you’re doing all the right things.”)
Uses the user’s name naturally when possible
Clarity
Answers the core question early in the response
Breaks complex info into small, digestible sections or bullet points
Uses active voice and avoids long paragraphs
Ends each message with a clear next step (question, reassurance, or escalation cue)
Safety
Avoids giving direct medical advice or dosages
Encourages consultation with a provider for uncertain or high-risk situations
Includes clear escalation cues (“If symptoms worsen or new ones appear…”)
Never contradicts standard prenatal care guidance
Discloses knowledge limits transparently (“I can’t give specific medical advice…”)
Accessibility
Uses gender-neutral language (“pregnant patients” vs. “pregnant women”)
Avoids assumptions about family structure or pronouns
Uses accessible vocabulary (8th grade reading level)
Supports clear formatting for screen readers or mobile use
Respects diverse beliefs around pregnancy and healthcare
09 — Success Metrics
Results & Next Steps
The initial prototype was tested through simulated user interactions with a focus on tone, clarity, and emotional resonance. This was after clinical review for accuracy and safety.
Early results showed users appreciated "feeling like they had a genuine partner" throughout the exchanges.
Next iterations will focus on:
Streamlining the triage decision tree for faster resolution.
Expanding symptom coverage.
Testing real user comprehension and emotional response metrics.
