Verbally

Why I Built This Most mental health apps focus on tracking symptoms, journaling consistently, or following guided exercises. But here's what I kept noticing: the hardest part isn't keeping up with an app. It's admitting something specific that you already know but won't say.

There's a moment that happens before someone seeks help, before they open up to a friend, before they even journal about it—when they're circling around a thought, rephrasing it, softening it, talking around it. That moment of avoidance is where clarity actually begins, but almost no tools are designed for it.

Verbally inverts the entire model. Instead of asking people to track more, write more, or engage more, it does one thing: identifies what they're avoiding and asks them to say it out loud. Once. Alone. That's where the intervention happens—not in the app, but in the private acknowledgment that follows.

What It Actually Does You write whatever's on your mind—a paragraph, a few sentences, whatever you're willing to type. Verbally analyzes it using an LLM to detect patterns of minimization, deflection, and linguistic avoidance. It doesn't look for keywords or diagnoses. It looks for what's unsaid.

Then it generates one carefully phrased sentence and gives you a single instruction: say this out loud, privately, once.

No follow-up questions. No mood tracker. No advice column. The entire impact happens outside the app, in that split second when you hear yourself say the thing you've been dodging.

This approach is radically different from existing mental health tools. It's not about engagement metrics or retention. It's about designing for a single moment of clarity that you'll never see—but that might change everything.

How It Works/How we Built It Verbally is built as a lightweight web application powered by Groq's API. The system architecture is as follows:

Input Analysis: The LLM processes written input to identify avoidance patterns—hedging language, deflection, passive constructions, and emotional minimization. Prioritization Engine: The AI ranks multiple potential "unsaid thoughts" and selects the single most avoided idea. Output Generation: A carefully phrased sentence is generated, optimized for spoken delivery rather than written reflection. No Data Retention: Nothing is stored. No tracking, no profiles, no history. Each interaction is isolated and private. How It Works/How We Built It

Verbally is a Next.js app that uses Groq's llama-3.3-70b model to detect what you're avoiding saying. It analyzes your text for linguistic avoidance patterns—hedging language, deflection, passive voice, minimization—then generates a single sentence you're instructed to speak aloud privately. After you confirm you've said it, the AI provides brief therapeutic guidance and asks if this truly captures what you're feeling, allowing you to iterate deeper. No data is stored. Built with React, TypeScript, Tailwind, Framer Motion, and Groq API—entirely stateless and private.

The Hard Part The biggest challenge was restraint. Every instinct said to add more features—journaling tools, progress tracking, follow-up prompts, community features. But every addition weakened the core idea.

The real technical challenge was tuning the AI to detect avoidance across diverse input styles. Someone who writes three formal paragraphs avoids differently than someone who writes two fragmented sentences. The system needed to work for both without becoming a generic sentiment analyzer.

We also had to solve for edge cases: What if someone writes something genuinely ambiguous? What if there's nothing being avoided? The AI needed guardrails to avoid false positives while still being decisive enough to generate meaningful output.

And then there was the design challenge of making the interface feel safe and non-judgmental. People are sharing vulnerable thoughts. The UI, copy, and pacing needed to communicate that this tool respects their autonomy and won't pathologize them.

What I'm Proud Of Verbally demonstrates a fundamentally new use case for AI in mental health. It's not diagnosis, it's not therapy, it's not mood tracking. It's insight surfacing—using AI to help people see what they already know but are actively avoiding.

The product is:

Culturally aware: It doesn't assume a clinical framework. It works whether someone's struggling with burnout, relationship issues, grief, or identity. Deeply empathetic: The AI is tuned to be direct but not harsh, specific but not invasive. Accessible: No signup, no subscription, no data collection. You use it when you need it. Grounded in research: Avoidance is a well-documented psychological mechanism. ACT (Acceptance and Commitment Therapy) and exposure-based therapies already validate the power of acknowledgment. Verbally just makes that first step radically easier. This isn't a feature-heavy platform. It's a precision tool that does one thing exceptionally well.

What I Figured Out AI is most valuable in mental health when it prioritizes insight over instruction. Telling someone what they're avoiding can be more powerful than telling them what to do.

I also learned that designing for offline impact is a completely different paradigm. Most apps optimize for engagement, retention, session length. Verbally optimizes for a moment that happens after you close the tab. That required rethinking every assumption about what makes a "successful" product.

Finally, I learned that simplicity is hard. It's easy to build more. It's much harder to cut everything that isn't essential and trust that what remains is enough.

What's Next Short-term:

Contextual refinement: Improve the AI's ability to detect avoidance across longer timelines and different input styles (formal vs. fragmented, abstract vs. specific). Multilingual support: Avoidance patterns differ across languages. Expanding beyond English requires cultural and linguistic recalibration. Optional reflection tools: For users who want to track patterns over time, we'll add a lightweight, opt-in reflection log—but it will always be secondary to the core experience. Long-term:

Integration pathways: Partner with therapists and mental health platforms to position Verbally as a pre-therapy tool—something people use before they're ready to seek formal help. Research validation: Conduct user studies to measure impact. Does saying something out loud actually create behavior change? How does this compare to traditional journaling? Accessibility features: Add voice input for users who struggle with typing, and optimize for screen readers. The goal is to keep Verbally minimal, intentional, and focused on the moment where acknowledgment begins. It will never become a full mental health platform. It's a tool for one specific, critical moment—and that's what makes it work.

Why This Matters Mental health tools have become synonymous with complexity. Apps with dozens of features, daily check-ins, gamification, community forums. But sometimes what someone needs isn't more engagement. It's permission to stop avoiding.

Verbally proves that you can build something deeply impactful without complexity. One AI model, one interaction, one sentence. That's it. And that might be exactly what someone needs to finally say the thing they've been carrying alone.

Built With

• framer-motion
• groq
• next.js
• react
• tailwind
• typescript
• vercel