CIRCA

Between a positive test and a first appointment, there’s almost no reliable support

Most women can’t see a doctor until around week 12. In that limbo, worry builds and reliable answers are nowhere to be found.

Penn Medicine built CIRCA to fill the gap — but patients weren’t signing up

CIRCA is a text-based support service from Penn Medicine’s PEACE clinic, designed to give women reliable, clinician-backed answers during those early weeks through an AI chatbot. But it wasn’t landing. The challenge wasn’t just improving CIRCA’s answers — it was making the service feel trustworthy enough to try in the first place.

CIRCA offers 24/7 access to clinically approved answers — a real gap-filler for women navigating early pregnancy without a doctor to call.

But the experience had significant friction. Due to NLP limitations, interactions felt robotic: responses were repetitive, inputs were rigid (e.g., “Y” or “N”), and there was no natural conversation flow. More critically, CIRCA lacked emotional support — leaving users feeling unseen during vulnerable moments. Occasional account deactivations added to the frustration.

Actual CIRCA User Screenshots

Helpful response

Repetitive messaging

Account deactivated

We approached CIRCA’s redesign as a trust problem, not a feature problem.

The existing service was clinically sound, but emotionally distant, and adoption remained low among the patients it was built to support.

User Interviews

12in-depth conversations to understand emotional and support needs

User Surveys

50+quick, diverse responses on AI tone preferences and support needs

Secondary Research

Studied how to make AI and design feel more human and trustworthy

Digital Ethnography

20+forums and apps studied to observe real pregnancy questions and behaviors

I helped translate research around trust, anxiety, and information overload into CIRCA’s service strategy. My work focused on synthesizing user needs, defining the adaptive tone model, rewriting onboarding to feel more trustworthy, mapping context-based care flows, and defining a pre-enrollment chatbot preview to make CIRCA more accessible and easier to trust before sign-up.

Behind every symptom question is an emotion that also needs an answer

When women reach out with a question, they’re rarely just after medical facts. They want reassurance and validation too. The best answers carry both — the information and the emotional support.

We gave CIRCA a voice that adapts to what’s being asked

AI personaRespectful, supportive, calming, compassionate, and clinically grounded — each attribute maps to a trust barrier from research, balancing non-judgmental warmth with Penn Medicine's clinical authority.

Hybrid tone modelCIRCA shifts its tone to match the question — calm and factual for medical ones, warm and validating for emotional ones.

Improved NLP with ChatGPT APICIRCA’s answers come from Memora Health’s databank of clinically approved responses. We added a ChatGPT layer that adjusts the language — keeping the medical accuracy intact while making each answer feel helpful and supportive rather than clinical.

Before

After

Same clinically-approved answers, delivered with warmth.

A one-size-fits-all flow couldn’t adapt to wanted, unsure, or loss

No two pregnancies are the same — even for the same person the second time around. A single linear flow couldn’t meet users whose experiences ranged from excitement to uncertainty to loss. Each pregnancy journey can look drastically different.

We built paths that match each person’s situation

Redesigned workflowDifferent approaches based on pregnancy preference.

Redesigned Workflow

The redesigned flow routes users at a single decision point — their pregnancy preference — captured right after they confirm the pregnancy.

Desired and unsure paths: Branches into tailored support for users who are excited or uncertain, instead of one linear flow for everyone.

Pregnancy loss support: Adds a dedicated path for early pregnancy loss, where the original flow had none.

The adaptive flow also needed to handle moments where support alone wasn’t enough.

Red-flag alertsDetects high-risk language like thoughts of self-harm, responds with empathy, and routes the person to a human care specialist instead of an automated reply.

User is acknowledged, validated, and safely escalated to professional help.

The original onboarding buried the service name, skipped any emotional welcome, and made no attempt to earn trust before asking for consent.

Streamlined onboardingReplaces the long, cold, spammy intro with a warm, plain-language welcome that names CIRCA, notes it’s backed by Penn Medicine, and states upfront it’s not for emergencies.

Before

Long, clinical, impersonal. Didn’t name the service or explain why to trust it.

Proposed

Warm, clear, and honest. Builds trust before the first exchange.

Trust starts with being found, and CIRCA was invisible

CIRCA’s outreach poster didn’t convey what the service was for, and online it had almost no presence. Users couldn’t trust — or even find — a service they didn’t know existed.

We made CIRCA easy to find, understand, and try

Poster redesignLeads with what CIRCA offers, building trust before the ask.

Before

No service name, no Penn Medicine branding, no clarity on what happens after you enroll. Generic imagery makes the purpose ambiguous, and the cold clinical palette signals authority without warmth — the wrong emotional register for an anxious first-time patient.

Proposed

A warmer color scheme and friendly tone make the service feel approachable. A clear program description, relevant visuals, Penn Medicine branding for credibility, and a link for more information give patients everything they need to trust and act.

We designed a layered discovery path — find CIRCA, understand it, try it, trust it.

Website redesignWarmer, more welcoming PEACE site that leads with patient needs, not clinic credentials.

CIRCA landing pageA dedicated page to learn what CIRCA is, what it does, and who it’s backed by — before enrolling.

Try before signing upAn embedded chatbot preview lets users experience CIRCA’s voice and helpfulness before committing.

Transparent data handlingA plain-language privacy page that addresses the data concerns users raised directly in research.

We prioritized changes by feasibility and trust impact across three phases — immediate outreach fixes, near-term UX improvements, and long-term research investments.

CIRCA changed how I think about AI in sensitive healthcare moments: a technically correct answer isn’t always a supportive one. Trust had to be designed across the whole experience — the first message, the tone of each reply, the ability to escalate risk, and how users found CIRCA in the first place. The biggest lesson was that AI here shouldn’t only optimize for accuracy; it has to communicate its role clearly, respect emotional context, and know when to step aside for human care.