Inspiration
When someone experiences a panic attack, overwhelming anxiety, or intrusive thoughts, even simple decisions can feel impossible. Searching online for coping strategies isn't particularly helpful when you're struggling to breathe, think clearly, or focus on what's happening around you.
And the problem doesn't necessarily end when that moment passes. Finding a therapist who understands your specific needs, accepts your insurance, and feels like the right fit can be an overwhelming process of its own.
We wanted to bridge the gap between needing help right now and finding meaningful support in the long term.
That's why we built Anchor, a voice-first mental health companion designed to meet people where they are, whether they're experiencing an overwhelming moment, trying to understand patterns in their mental health, or looking for professional care.
What it does
Anchor combines immediate, voice-based emotional support with long-term mental health tracking and personalized care discovery.
Talk through difficult moments: When someone is experiencing panic, anxiety, intrusive thoughts, or overwhelm, they can speak directly with Juno, our ElevenLabs-powered voice companion. Instead of presenting a wall of text, Juno guides users through appropriate grounding exercises and coping strategies, adapting to what they're experiencing. Users can interrupt naturally, just as they would in a real conversation.
Find guidance grounded in reliable sources: Using Bright Data, Anchor retrieves information from trusted mental health resources, including the NHS, NIMH, and Mayo Clinic. Gemini processes this information into practical, conversational guidance, helping users understand not just what to do, but why a particular technique might help.
Keep a conversational diary: Users can speak or write about their day and receive thoughtful follow-up questions and reflections. Anchor helps them organize their thoughts, recognize recurring concerns, and identify things they might want to discuss with a therapist.
Track mental health over time: Anchor lets users monitor panic episodes, OCD-related urges, and daily mood, helping them identify patterns that might otherwise go unnoticed.
Find professional support: Rather than relying on a static directory, Anchor uses Bright Data to discover publicly available information about therapists and mental health services. Users can search by location, insurance, and type of care, making it easier to find relevant options. Listings are explicitly marked as unverified.
Stay on top of therapy: For users already receiving treatment, Anchor helps manage appointments, track attendance, organize therapy homework, and prepare topics for upcoming sessions.
Adapt to individual needs: Anchor offers customizable interfaces, including differences in spacing, motion, colors, and information density, to make the experience more comfortable for different users.
Anchor also includes safety checks that identify potentially concerning situations and direct users toward appropriate crisis or emergency resources. It is designed as a supportive tool, not a replacement for professional mental health care.
How we built it
We built Anchor using Next.js, TypeScript, ElevenLabs, Gemini, Bright Data, and Neon, with each technology handling a different part of the experience.
ElevenLabs — Making support conversational
We integrated ElevenLabs to give Anchor a natural, responsive voice. Our voice interface supports interruptions, allowing users to speak over the agent without waiting for it to finish. We also implemented voice personalization and fallback audio to make interactions more accessible and reliable.
Gemini — Understanding and responding
Gemini powers conversational intelligence, diary reflections, safety classification, and the interpretation of retrieved mental health information. It helps transform complex guidance into simple, actionable steps that can be delivered through natural conversation.
Bright Data — Connecting users to real-world information
We used Bright Data for two major purposes:
- Retrieving publicly available guidance from trusted mental health websites to inform coping strategies.
- Discovering mental health professionals and care options based on users' locations, insurance, and care requirements.
We also used Bright Data to research accessibility-oriented interface design guidance, which Gemini translates into personalized layout settings.
Neon — Making support continuous
Neon serves as our persistent database, storing user preferences, check-ins, diary entries, therapy plans, mental health tracking data, and retrieved care information. This allows Anchor to connect individual interactions into a longer-term support experience.
Next.js and TypeScript — Bringing everything together
We built the application around an accessible interface with dedicated sections for immediate support, journaling, tracking, therapy management, and care discovery.
For immediate guidance, we implemented a structured conversation flow with safety checks, streamed speech, and fallback responses, prioritizing responsiveness and reliability.
Challenges we ran into
Making voice interactions feel natural
During moments of anxiety, even a small delay or an overly long response can make an interaction frustrating. We had to think carefully about latency, interruptions, and how to guide someone without overwhelming them with information. Supporting speech interruptions while maintaining the flow of a conversation was a particularly interesting technical challenge.
Balancing personalization with safety
Mental health support isn't one-size-fits-all. Someone experiencing a panic attack may need a very different interaction from someone dealing with intrusive thoughts. We needed to make guidance context-sensitive without allowing an AI model to improvise potentially inappropriate advice.
To address this, we combined structured guidance with safety rules, reviewed exercises, and Gemini-based classification.
Turning web research into usable guidance
Retrieving information is relatively straightforward. Turning that information into short, understandable instructions for someone in distress is much harder.
We had to connect Bright Data's retrieval capabilities with Gemini's ability to interpret information, while keeping track of sources and ensuring the guidance remained appropriate.
Making care discovery genuinely useful
Finding therapists involves more than searching for nearby providers. Insurance coverage, specialization, and personal preferences all matter. We worked on transforming unstructured public information into searchable provider listings while acknowledging that online information may be incomplete or outdated.
Accomplishments that we're proud of
Building an end-to-end mental health support experience: Anchor doesn't stop at providing a chatbot. It connects immediate support, ongoing reflection, mental health tracking, therapy management, and professional care discovery in one application.
Creating an interruptible voice companion: We built a conversational experience that lets users speak naturally, interrupt responses, and receive guidance without navigating complicated menus.
Integrating four major technologies meaningfully: ElevenLabs enables voice interaction, Gemini provides intelligence and interpretation, Bright Data connects users to current public resources, and Neon supports continuity across sessions.
Prioritizing safety and privacy: We implemented safeguards for concerning situations, persistent access to crisis resources, optional transcript retention, and controls for deleting personal data.
Designing for accessibility: Rather than treating accessibility as an afterthought, we incorporated customizable layouts and researched interface preferences into the product itself.
Most importantly, we're proud that Anchor addresses both sides of a difficult problem: helping someone navigate an overwhelming moment and making it easier to access ongoing support afterward.
What we learned
One of our biggest takeaways was that building AI for mental health requires a very different approach from building a typical AI assistant.
A technically impressive response isn't necessarily a helpful one. When someone is panicking, the most useful response might be a single, clearly spoken instruction rather than a detailed explanation.
We also learned the importance of combining different kinds of technology rather than expecting one model to solve everything. Voice models make interaction accessible, language models help interpret context, real-time retrieval connects responses to external information, and persistent databases help turn isolated conversations into a more continuous experience.
Finally, we learned that AI works best in sensitive applications when its limitations are explicitly designed around. Safety rules, user control, source transparency, and access to human support are just as important as the underlying intelligence.
What's next for Anchor
We see Anchor evolving into a more comprehensive, personalized mental health support platform.
Our next priorities include:
- More personalized support: Improving how Anchor adapts its conversational style and guidance to individual preferences, past interactions, and recurring patterns.
- Better therapist matching: Expanding care discovery to consider specialties, therapeutic approaches, insurance compatibility, availability, and user preferences.
- Improved accessibility: Developing more flexible interaction modes, including additional languages and voice options.
- Stronger continuity of care: Helping users turn their journal entries, tracked symptoms, and reflections into useful summaries they can choose to share with mental health professionals.
- Clinical validation: Working with qualified mental health professionals to evaluate the appropriateness, safety, and effectiveness of the experience before any real-world clinical deployment.
Our long-term vision is for Anchor to become a reliable first point of support: something people can turn to when they feel overwhelmed, and something that helps them take the next step toward meaningful, human care.
Built With
- ai
- bright-data
- elevenlabs
- gemini
- javascript
- llm
- mental-health
- neon
- next.js
- postgresql
- react
- tailwind-css
- typescript
- voice-ai
- web-scraping
