WELLness - Mind ML

A lightweight, fully local sentiment analysis web application designed to assess user mental state based on their typed input — without relying on any external APIs or hosted ML models.

Comment 1

Inspiration

The rising mental health crisis around the world, especially among youth, inspired us to build something meaningful yet lightweight — a wellness checker that respects privacy and runs real ML, built from scratch, with no LLMs or APIs.

We also took this hackathon’s theme seriously — "Build real ML web apps with no wrappers" — and wanted to prove that small yet impactful ML apps can be built and deployed using your own trained models.

What it does

  • Takes user input related to their thoughts or feelings
  • Classifies sentiment (Positive, Neutral, Negative) using a trained ML model
  • Displays emoji-based feedback and results in real-time
  • All processing is done locally using a custom-trained model — no external APIs involved

How we built it

  • Built a Logistic Regression model using Scikit-learn on a curated dataset of text sentiments
  • Converted the model to a .pkl file and integrated it with Flask for backend inference
  • Developed a single-page responsive HTML/CSS/JS frontend
  • Sent data via AJAX to predict user sentiment and return a response
  • Ensured minimal dependencies and total local model logic, with zero wrappers or cloud-based ML tools

Dataset Used : https://www.kaggle.com/datasets/suchintikasarkar/sentiment-analysis-for-mental-health

Google Colab Notebook : https://colab.research.google.com/drive/11j8Obtap3SAMtm8K896imJbn_WJ_YjDo?usp=sharing

Challenges we ran into

  • Tuning model accuracy with limited data
  • Seamlessly integrating the frontend with Flask using vanilla JS (no React or complex frameworks)
  • Keeping the app lightweight, responsive, and fast without using any hosted model APIs
  • Designing a clean UI that feels modern and accessible despite time limits ## Accomplishments that we're proud of
  • Built an end-to-end ML web app from scratch, no wrappers
  • Achieved over 85% model accuracy
  • Made mental health checking accessible and private
  • Delivered a polished, one-page experience that works across devices

What we learned

  • How to train and save models efficiently for real-time inference
  • Flask + JS integration tricks for smoother UX
  • Importance of keeping UX minimal for wellness/mental health tools
  • How powerful even small models can be when paired with thoughtful design ## What's next for WELLness - Mind ML
  • Add mood tracking over time with graphs and history
  • Build a mobile-first PWA version
  • Partner with counselors/NGOs to offer anonymous feedback
  • Train model on more nuanced emotion classes like anxiety, stress, joy, confidence

Built With

Share this project:

Updates