Luna. AI Clinical Assistant
Mental health practitioners are under constant pressure not just emotionally but logistically. They juggle session notes, assessments like the GAD-7, documents from clients and their own evolving observations about patient progress. Much of this critical information is spread across notebooks, spreadsheets and memory, making it difficult to track changes over time or confidently tailor treatment plans.
“I know the data is there but pulling it together while juggling sessions is just overwhelming.”
Enter Luna
Luna is an AI-powered assistant designed to work alongside clinicians, not instead of them. It helps collect, synthesize, and surface relevant patient insights from session notes, assessments and uploaded documents transforming scattered information into something clear, accessible and actionable.
More than just a data organizer, Luna also makes smart, evidence-based recommendations that evolve as new data is added. The goal is to reduce admin load and improve clinical decision-making without ever getting in the way of the human connection between practitioner and patient.
Luna doesn’t just summarize, it adapts. As new data is added, its understanding of the patient deepens and its recommendations evolve. It’s a tool designed to support human care, not replace it.
The interface merges AI-driven voice sentiment analysis with clear, human-readable insights.
Visual Hierarchy: Immediate summary tone, clinical insights and recommendations are surfaced top-left for rapid comprehension.
Emotional Timeline Table: Condensed and confidence-rated signal breakdown aligns with clinician workflows.
Customizable Templates: Users can choose from pre-made templates or create their own for LUNA to structure session recordings, ensuring flexibility and personalization in documentation workflows.
Designing with (and for) Clinicians
My role as UX designer and researcher was to deeply understand the day-to-day lives of clinicians and ensure Luna fit seamlessly into their existing workflows.
I started with exploratory research: interviews with mental health professionals across different settings. They shared how much time they spend summarizing information, how easily things fall through the cracks, and their mixed feelings about AI in therapy.
They didn’t want AI to replace their judgment but they did want support.
“One psychologist told me, ‘If it can surface the right info at the right time, that’s actually huge.’ That became a guiding design principle.”
The Experience
Luna is built around a three-panel interface that mirrors how clinicians think.
On the left, the Patient Profile shows clinical scores like GAD-7 and structured patient information: history, presenting issues, family background, lifestyle factors. Each of these items acts as a smart trigger, clicking on one sends a prewritten prompt to Luna, asking for its current understanding. Alongside each item is a circular confidence indicator. If it’s full, Luna has enough data to give strong insights.
The center panel is where the conversation with Luna happens. Clinicians can interact with the AI in two ways: by clicking prompts from the left panel or typing free-text questions like, “How has anxiety changed over the last month?” Luna responds with a concise summary based on everything it knows by pulling from notes, assessments and uploads. It also suggests follow-up actions or additional questions to explore and can generate graphs when relevant, such as symptom trends over time.
“I love that I can see what Luna knows and what it doesn’t at a glance.”
On the right, the Recommendations panel provides a constantly evolving summary of Luna’s suggestions for care. This might include next actions, like “Re-administer GAD-7 next session” or higher-level recommendations based on symptom patterns. There’s also a “Full Report” button that opens a detailed, clinician-friendly summary with rationale and treatment planning.
The interface uses accordions and tooltips to progressively reveal or hide information based on the clinician’s needs. This design approach minimizes cognitive load, supports focused decision-making, and ensures that detailed insights like score breakdowns or reasoning behind recommendations are accessible on demand without cluttering the main view.
Clean, organized clinical interface showing patient profile, assessment scores, emotional functioning data and AI-generated treatment suggestions with expandable details for deeper analysis.
The clinician maintains full control over every AI-generated recommendation. As shown in the image, they can override any section directly or use the “Ask LUNA to edit” feature to revise the output. This ensures that clinical judgment and personalized care always take precedence.
Using Luna: A Realistic Flow
Imagine a clinician just finished a session. They open the patient’s profile to check recent GAD-7 scores, then click on “Presenting Issues.” Luna instantly replies in the chat with its summary, along with a list of what data is missing. The clinician asks a follow-up question, “Is there any sign of a relapse?” and Luna offers a chart of anxiety levels over time and a nuanced answer based on past sessions. As a result, the Recommendations panel updates with a focused suggestion for next steps.
The clinician can then click into the full report, download it, or use it to prep for their next conversation. Everything is in one place, without digging through disconnected files.
This clinician interface integrates emotional functioning domains, session-by-session GAD-7 and PHQ-9 visualizations and interactive AI prompts. Users can explore changes over time, toggle between psychological dimensions and chat with LUNA for context-aware insights.
“It’s like having a colleague who never forgets anything.”
What We Heard from Users
In usability tests, clinicians found the experience intuitive and genuinely helpful. They liked seeing what Luna “knew,” and appreciated being able to use it like a sounding board—not just a search tool. The confidence indicators made it easy to see where to focus and perhaps most importantly, they felt Luna didn’t try to take over—it simply supported their work.
Their feedback helped refine the tone of Luna’s suggestions, add direct “quick actions” from replies (like “Upload now” or “Ask in next session”), and make the chat feel more conversational.
Design Takeaways
This project challenged me to think about AI not just as a tool, but as a presence in a sensitive professional space. Luna needed to be intelligent, but also respectful—transparent, but not overwhelming. Balancing power with subtlety was key.
It also reminded me of the power of structured and flexible interaction. Clinicians liked clicking prompts when they were in a rush, but also appreciated being able to ask Luna anything in natural language.
