SerenityAI

Inspiration

In today’s fast-paced world, anxiety, burnout, and emotional fatigue are common, especially among students and working professionals. We realized that while there are AI tools for productivity, there’s a significant gap in accessible, supportive mental wellness tools — particularly ones that feel emotionally intelligent, empathetic, and human-centered.

We were inspired to create SerenityAI as a safe space — a calm, conversational AI therapist built using open-source models — for users to express themselves, receive emotional support, and get motivation when they need it most.

What it does

SerenityAI is a web-based mental wellness companion powered by LLaMA 3 (via Groq API). It provides: 🧘 Chat with an empathetic AI therapist: Users can talk to the AI and receive emotionally aware, calming responses. 💬 Mood detection with emoji feedback: Based on what the user writes, the AI detects the emotional tone and reflects it visually using emojis.

Every response from SerenityAI is grounded in kindness, clarity, and compassion — just like you'd want from a human counselor.

How we built it

💻 Frontend: Built with React.js (Vite), styled with custom external CSS for a sleek, responsive UI. We designed a calming, dark-themed experience to enhance focus and reduce screen fatigue. ⚙️ Backend: Powered by FastAPI (Python) with endpoints for chat, affirmation, motivation, and wellness tips. 🧠 AI Model: We used Groq’s blazing-fast LLaMA 3 (8B) model for generating emotionally sensitive responses, affirmations, and motivational content. 📦 Environment Variables: Managed API keys securely with python-dotenv. 🌐 Communication: Used fetch API to connect frontend and backend. 💡 Extras: Mood detection logic was custom-crafted to interpret emotion from text and return appropriate emojis.

Challenges we ran into

⚠️ Tuning LLaMA 3 for emotion-aware replies was tricky. Some prompts produced generic or robotic responses, so we spent time engineering prompts that feel more “human.” 💥 Initially got repeated 404 errors from the Groq API due to small mistakes in model names and payload formatting. 🎯 We had only a few hours to build the entire project, so we had to carefully prioritize must-have features over ambitious extras. 💬 Handling natural language inputs and balancing empathy with consistency was a major design challenge.

Accomplishments that we're proud of

✨ Created a polished, minimal mental wellness assistant in under 6 hours. 🤖 Successfully integrated an open-source LLM (LLaMA 3) via Groq to deliver calm, context-aware emotional responses. 💡 Engineered mood detection logic from scratch — with no external libraries — to make the app feel emotionally intelligent. 🎨 Designed a user interface that is clean, calming, and visually engaging — a great user experience even under time constraints. 🚀 Deployed a fully working demo with proper routing, component structure, and future scalability.

What we learned

👥 Empathy in design is just as important as technical functionality — the tone and feel of the AI responses really matter. ⚙️ Learned how to integrate and debug Groq API effectively, including API key management, headers, and prompt tuning. 🎨 Understood how simple UI decisions (like emoji feedback or a darker background) can make an app feel significantly more user-friendly and emotionally aligned. 🧠 Learned how to write better prompts for LLMs to generate more coherent and on-topic outputs. 🛠️ Realized the importance of modular backend structure (one file per route) for future scalability.

What's next for SerenityAI

📈 Emotion tracking over time: Let users view their mood trends with charts and personal journals. 🎙️ Voice input & response: For more natural, hands-free conversations. 💌 Email-based daily affirmations & reminders 🌍 Multi-language support using translation APIs 🧠 Self-care plan generator: Based on the user's mental state and mood patterns 🔐 Add optional user accounts and data privacy safeguards for returning users

Built With

• and-custom-css
• fastapi
• fetch
• groq-api-(llama-3)
• javascript
• python
• react-(vite.js)