Home Page
AI Companion Chat
SOS(Emergency Button)
SOS Options
SOSFunction
Journal

🌸 Bloom: AI Mental Wellness Companion

Inspiration 💡

Mental health support is often inaccessible, expensive, or intimidating. We wanted to bridge this gap by creating Bloom—an AI companion that's always available, judgment-free, and psychologically safe. The goal wasn't to replace therapy, but to provide that crucial "daily maintenance" layer of emotional support. We were inspired by how large language models can simulate empathy, but we noticed existing tools often felt robotic or lacked robust safety rails. We wanted to build a sanctuary where users could feel truly heard using Gemini 3's advanced reasoning capabilities.

What it does 🤖

Bloom is an AI-powered wellness companion that provides a safe space for emotional processing. Unlike generic chatbots, Bloom is purpose-built for mental wellness with four core pillars:

❤️ Empathetic Chat: A warm, supportive companion that uses active listening and cognitive reframing techniques to help users navigate their emotions.
🌬️ Wellness Tools: Interactive, guided exercises like Box Breathing, 5-4-3-2-1 Grounding, and Body Scans to help manage acute anxiety or stress.
📊 Mood Tracking: A judgment-free space to log feelings and visualize emotional trends over time.
🚨 Crisis Intervention: Built-in safety protocols that detect distress signals and immediately provide professional resources (like 988) while maintaining a supportive presence. ## How we built it 🛠️ We built Bloom using Google AI Studio, leveraging the Gemini 3 model family.
System Architecture: We designed a comprehensive system prompt that functions as the app's "brain." This prompt encodes therapeutic best practices, safety boundaries, and personality traits directly into the model's context.
Prompt Engineering: We utilized refined system instructions to handle complex behaviors—switching seamlessly between open-ended conversation and structured exercises (like counting breath cycles).
Gemini 3 Features: We utilized Gemini 3's improved reasoning to detect nuance in user sentiment, allowing Bloom to distinguish between "venting" and "crisis," and to offer more context-aware responses than previous generations. ## Challenges we ran into 🏔️
Balancing Empathy & Safety: It was challenging to make Bloom feel warm and close without crossing professional boundaries. We had to carefully fine-tune the prompt to ensure it never attempts to "diagnose" or "treat" while still validating feelings.
Structured Outputs in Chat: Getting a text-based model to effectively lead a real-time breathing exercise (pacing the counts) required multiple iterations of instruction tuning in AI Studio.
Crisis Detection: Ensuring the model strictly adheres to safety protocols when it detects self-harm keywords, regardless of the conversation flow, was our top priority and required rigorous testing. ## Accomplishments that we're proud of 🏆
Human-Like Connection: We're proud of Bloom's "personality." It genuinely feels warm and grounded, not helpful-but-cold like a standard assistant.
Integrated Safety Nets: Successfully implementing a robust safety protocol that feels supportive rather than punitive or robotic when a user is in distress.
All-in-One Experience: Successfully combining chat, tools, and tracking into a single cohesive experience using AI Studio's app builder capabilities. ## What we learned 📚
The Power of Context: We learned how much "personality" and domain expertise (like therapeutic techniques) can be encoded simply through high-quality system prompting with Gemini 3.
Safety First: We deepened our understanding of responsible AI development, specifically why distinct guardrails are essential for health-related applications.
Gemini 3's Reasoning: We were impressed by how well Gemini 3 could "read between the lines" of a user's message to identify the underlying emotion (e.g., identifying fear behind anger). ## What's next for Bloom 🚀
Mobile App Integration: Taking the core logic validated in AI Studio and deploying it as a TypeScript and the React (which we have already started prototyping!). For future development we will convert this to flutter for cross-platform deployment
Personalized Voice Mode: Implementing low-latency voice interactions for real-time talk therapy sessions.
Long-Term Memory: Implementing a secure way for Bloom to remember past conversations to provide deeper, more continuous support over weeks and months.