Lumi - Mental Health Companion

👥 Team Members

Name	GitHub Handle	Contribution
Maya Swan	@msmayaswan	Built Sentiment NLP model, designed wellness exercises, assisted with frontend development
Jolisa Fields	@Jolisa-dot	Backend Integration, Developed core backend features including the journal system, daily check-in system, and the mood trend graph. Integrated MongoDB for data storage, built RESTful API routes
Dana Brunson	@danabrunson	Frontend design, frontend and backend feature integration
Laila Donaldson	@lailacodes	Designing UI utilizing Figma, assisting with the development of frontend design and integration
Nicolas Harris	@NicolasHarris	Backend

---

🎯 Project Highlights

• Lumi analyzes the environment and uses information provided by the user to supply resources/exercises suited to each individual.

• Using NLP to interpret natural language for sentiment analysis.

• Implemented AR/3D/2D tools to optimize user experience through AR exercises.

• This project defines different techniques such as machine learning, UI, and backend development.

---

👩🏽‍💻 Setup & Execution(Backend)

Step 1: Clone the Repository Open your terminal

Run: git clone https://github.com/Jolisa-dot/StackUnderflow-SeniorProject.git cd StackUnderflow-SeniorProject/Lumi/Code

Step 2: Install Backend Dependencies Inside the Code/ folder, install all Node.js dependencies:

npm install

Step 3: Set Up Your Environment File In the same Code/ folder, create a file named .env

Add the following content:

MONGO_URI=mongodb+srv://<your_username>:<your_password>@cluster.mongodb.net/lumi?retryWrites=true&w=majority PORT=5000

Step 4: Start the Server From the Code/ folder, start the backend server:

node server.js This will run your Express server on http://localhost:5000.

Step 5: Test the Features You can now access and test your backend:

If you're using HTML files like journal1-frontend.html or dailycheckin.html, open them in your browser

---

🏗️ Project Overview

Lumi is a web application built using React.js and machine learning-powered sentiment analysis, designed to provide personalized emotional support. It allows users to express how they feel through journaling and mood tracking, and receive AI-generated feedback via features such as:

*Personalized Daily Check Ins *Journaling & Reflection Feature *Mood Tracking *Resource Hub with External Links *AR/VR/2D based Coping Exercises *Sentiment NLP Text Analysis

Lumi promotes mindfulness and self-reflection by turning user input into actionable emotional insights, helping users build healthy mental wellness habits over time.

---

🧠 Model Development

We implemented two pre-trained transformer-based NLP models using Hugging Face:

• distilbert-base-uncased-finetuned-sst-2-english for binary sentiment analysis (positive/negative)

• emotion-english-distilroberta-base for multi-label emotion classification (e.g., joy, sadness, fear, anger)

Training Setup: The models were not trained or validated on custom data within our project scope. All journal and mood input was processed via real-time API calls on user-submitted text. Evaluation of model performance was done based on a qualitative review of model output compared to the expected emotional tone from test journal entries.

Evaluation Metric: Since our project focuses on in-app emotional response and flagging rather than classification accuracy, evaluation was based on:

Subjective alignment between detected emotion and intended tone Whether flagged harmful entries triggered the appropriate emergency resources

---

📈 Results & Key Findings

Overall Model Performance:

Sentiment detection performed consistently well on casual, emotional, and reflective journal entries. The emotion classification model provided multiple probable emotions with relative confidence scores, allowing nuanced reactions from the system.

🚀 Next Steps & Future Improvements

• Involves improving the application by providing security and frontend and backend improvements.
• Improving the user experience can be achieved through UI updates and more functionality on the application.
• Much data is generated, providing a space to improve and add additional AI features.
• Future aspirations: overall enhancement and/or mobile deployment.

---

📄 References & Additional Resources

*Currently, there are over 10,000 mental health and wellness applications available, each taking different approaches to supporting users’ well-being. Many focus on mood tracking, mindfulness, or cognitive behavioral therapy (CBT)-based techniques, but few provide a comprehensive, interactive, and adaptive experience that combines AI-driven insights and habit-building features.

*According to the National Institute of Mental Health, approximately one in five adults in the U.S. experience mental health challenges.

---