Somna: An AI companion for the nights when sleep won't come.

Overview

Most sleep apps track your sleep. They leave you alone with the data.

Somna is a mobile app for adults struggling with inconsistent sleep routines, overstimulation, or late-night anxiety. At the center is Luna, an AI companion who checks in, adapts to your actual life, and talks to you like a person, not a dashboard.

Many adults struggling with sleep don't need more data. They need support during the moments routines break down. That's the problem Somna is designed to solve.

Research & Thinking

I spent time with four existing sleep apps: Better Sleep, Headspace, Sleep Cycle, and Calm. I wasn't just looking at features. I was looking for where users lose momentum and why.

Across all four apps, one pattern kept surfacing: users were given plenty of information but very little support in the moment they needed it most. Every app focused on what happened after sleep, not the fragile transition before it. That transition is where behavior actually breaks down.

Four insights shaped everything that came after:

• Data creates distance, not change. Sleep scores explain what happened, but rarely help users act differently. When metrics feel confusing or inconsistent, users disengage.
• Some engagement loops unintentionally prolong wakefulness. Calm's podcasts and playlists kept users engaged right when they needed to wind down, contradicting the product's purpose.
• Routines assume consistency that most users don't have. Every app I looked at built around fixed schedules. For users with anxiety, ADHD, or unpredictable lives, a rigid routine is just another thing to fail at.
• AI without empathy is just a search bar. Sleep Cycle's bot Luma was limited to medical references and doctor lookups. She could surface information, but couldn't meet a user emotionally. That's the gap Somna exists to fill.

I kept coming back to one reference for Luna: a nurse. Not a doctor who diagnoses, not a coach who pushes. A nurse who shows up, asks the right questions, and makes you feel like things are going to be okay.

Design Strategy

Traditional sleep apps are static. They track patterns but can't adapt to changing routines or respond when a user is struggling at 2am.

Somna uses AI because sleep struggles are inconsistent by nature. A user's bedtime shifts. A bad week derails a routine. Anxiety spikes with no warning. A static app can log those moments. Luna can respond to them.

Luna adapts her check-ins to changing routines, offers support before users ask, and connects them to professional resources when needed, all without feeling intrusive or judgmental.

That's the difference between a product that tracks sleep and one that supports it.

Solution

Somna is designed around three flows.

Routine Building

Luna uses the morning alarm as an anchor and builds the wind down routine around it. Onboarding is conversational, not clinical. I anchored the routine to the morning alarm deliberately because it's the one consistent moment most users already have. Everything else builds around that single fixed point so the routine feels achievable, not prescribed.

In-the-Moment Support

Home is Luna's home base, a warm low pressure check-in every time the user opens the app. Luna Chat is where users go when they can't sleep. I designed Luna to listen first and respond with care before offering anything actionable. That order matters because leading with resources feels clinical. Leading with presence feels human.

Reflective Insights

Sleep Insights surfaces one plain language observation per week. Not a score, not a grade. I chose one insight deliberately because too many observations creates the same overwhelm that made existing sleep apps feel exhausting. One honest, specific thing the user can actually do something with.

Process & Tradeoffs

The clearest decisions I made were the ones where I said no.

I rejected daily check-ins because fixed notifications risk creating their own sleep anxiety. Luna checks in adaptively, and if a user goes quiet for several nights, she follows up once gently then waits for them to come back on their own terms.

I rejected a sleep score. Scores invite comparison and judgment, and a user who slept poorly doesn't need a number confirming it. Sleep Insights surfaces one honest, actionable observation instead, never a grade.

I made Luna conversation-first instead of resource-first. Articles and doctor references are available when needed, but they're never the opening move. Luna listens before she links.

Accessibility

I think about accessibility as a design baseline. With Somna, that meant two groups specifically: users with anxiety and users with ADHD.

For anxiety users, the app itself can become a stressor if it's not handled carefully. Luna never judges. She never makes a user feel like they failed. Three bad nights in a row and she checks in with curiosity, not concern.

For ADHD users, a rigid sleep schedule isn't realistic. My Routine was designed to flex around the user. Luna adapts when routines shift because for ADHD users, shifting is the routine.

Reflection & What's Next

If I could push Somna further, the feature I'm most excited about is community. Sleep struggles are isolating. There's something uniquely lonely about being awake at 2am when the rest of the world feels silent. I think Somna could create a quiet, low pressure space where users feel less alone in the dark.

Before building further, I'd test Luna's conversation flows with real users. The success of Somna depends on whether her tone feels supportive rather than intrusive, especially for users with anxiety, ADHD, or little experience with wellness apps. Small shifts in wording could dramatically change how the product is perceived.

What stuck with me throughout this project is that sleep struggles are lonely. People don't need a score. They need something that shows up. That's what I tried to design.