AI Wellness Hub

Inspiration

Workplace mental health is one of the most significant challenges of our time. While many companies want to help, they often lack the data-driven tools to know where to start. This leads to generic wellness programs that miss the mark. Our inspiration was to move beyond good intentions and build an intelligent tool that could provide concrete, actionable insights for organizations, while also offering direct, accessible support to any employee who needs it. We wanted to create a solution that was not just predictive, but genuinely helpful and human-centric.

What it does

The AI Wellness Hub is a comprehensive, multi-page application designed to foster mentally healthy workplaces. It features two core components: 📈 The Workplace Wellness Advisor: An enterprise-grade tool for managers and HR leaders. It analyzes a company's policies and culture to provide a suite of insights: Predictive Analysis: Forecasts the likelihood of employees needing mental health treatment. Wellness Report Card: Scores the workplace on "Support Structures" and "Open Culture." AI Explainability (SHAP): Reveals the exact factors driving the prediction, turning a "black box" AI into a transparent, trustworthy partner. Generative AI Companion (Gemini): Provides a detailed, empathetic analysis of the results with actionable recommendations. AI Action Kit: Instantly generates draft emails and meeting agendas to help managers act on the insights. Downloadable PDF Report: Creates a professional, shareable summary of the complete analysis. 💬 The Mental Healthcare Chatbot: A safe, anonymous, and public-facing resource for everyone. It uses Google's Gemini to provide reliable information on mental healthcare topics (e.g., "What is CBT?", "How do I find a therapist?"). It is strictly programmed to not give medical advice and to immediately provide crisis hotline numbers if a user appears to be in distress.

How we built it

Our development journey was an iterative process of building, testing, and intelligently improving our solution. V1: The Initial Model - We began by cleaning the OSMI Mental Health in Tech Survey dataset and training several classification models (XGBoost, SVM, RandomForest) to predict whether an employee had sought treatment. V2: The Critical Insight - We built a Streamlit application around our best model. However, we quickly discovered a critical flaw: the model was "lazy." It relied heavily on non-actionable features like Age and family_history, making the app unresponsive to the policy changes a manager would actually input. Although statistically accurate, it wasn't practically useful. V3: The Strategic Pivot - This led to our biggest breakthrough. We re-engineered our entire approach, creating a focused "Advisor Model" by intentionally removing the non-actionable features. This forced the AI to learn the more subtle patterns within workplace policies and culture, making the tool genuinely responsive and useful. V4: The Full-Featured Application - We then built the complete multi-page application in Streamlit, integrating a suite of advanced features: AI Explainability using the shap library to build trust and provide deeper insights. A Wellness Report Card and a downloadable PDF report using fpdf2 to provide tangible value. A Generative AI Companion and Action Kit powered by the Google Gemini API to provide detailed, human-like advice and practical tools. A dedicated, safety-focused informational chatbot. The entire application is built with Python, and all the code is organized for a multi-page Streamlit experience.

Challenges we ran into

We faced several real-world development challenges that we were proud to overcome: The "Lazy Model" Problem: Our biggest challenge was realizing our initial high-accuracy model was practically useless. Overcoming this required a complete strategic pivot, which ultimately led to a much better and more ethical project. Python Environment & Dependency Conflicts: Moving from a cloud notebook to a local environment and ensuring all library versions (scikit-learn, shap, fpdf2) were perfectly synchronized was a major technical hurdle. We solved this by creating a precise requirements.txt file. Streamlit Caching Errors: We encountered and fixed several advanced Streamlit errors, like the UnhashableParamError, by learning how to correctly structure our function calls for caching. Accomplishments that we're proud of We are incredibly proud of building a project that is not just a single model, but a complete, end-to-end system. Specifically: Successfully integrating a predictive AI (XGBoost) with a generative AI (Gemini) to create a "Dual-AI" solution. Going beyond prediction to implement AI explainability with SHAP, making our tool transparent and trustworthy. Building a beautiful, feature-rich, multi-page application that feels like a professional, polished product. Most importantly, creating a tool that has the real potential to help organizations build healthier and more supportive environments for their employees.

What we learned

This project was a masterclass in the difference between academic data science and real-world AI product development. We learned that the highest accuracy score isn't always the best metric for success. Usefulness, interpretability, and ethical design are far more important. We also learned the critical importance of a disciplined development process, from managing Python environments to building a user-friendly interface.

What's next for AI Wellness Hub

The AI Wellness Hub is built to grow. Our next steps include: Expanding the Dataset: Using the "Contribute" feature to collect more anonymous data and continuously retrain and improve our model. Long-Term Trend Analysis: Building a dashboard for companies to track their wellness scores over time after implementing changes. Multi-language Support: Expanding the chatbot and advisor to be accessible in multiple languages.

Built With

• fpdf2
• github
• google-gemini-api
• html/css
• pandas
• python
• scikit-learn
• shap
• streamlit
• xgboost