MentiFy is an AI-powered mental wellness web application that analyzes user responses using a fine-tuned DistilBERT model and generates personalized mental health insights with Google Gemini LLM. Through guided questions and a short behavioral assessment, the system provides users with an AI-assisted mental wellness report.
-
User answers 5 guided mental wellness questions.
-
Each response is analyzed using a fine-tuned DistilBERT model.
-
The model predicts one of 7 mental health categories for each response.
-
All responses and predictions are sent to Google Gemini LLM.
-
Gemini generates a personalized report including:
- Mental wellness summary
- Possible causes
- Personalized recommendations
- Healthy daily habits
- Risk level assessment
-
The user also completes a 10-question MCQ-based behavioral assessment for additional insights.
- HTML
- CSS
- JavaScript
- Flask
- REST APIs
- Fine-tuned DistilBERT
- PyTorch
- Hugging Face Transformers
- Scikit-learn
- NumPy
- Pandas
- Google Gemini API (
google-genai)
-
Architecture: DistilBertForSequenceClassification
-
Training Dataset: Kaggle Mental Health Dataset (~55,000 samples)
-
Accuracy: 87%
-
Classification Categories:
- Anxiety
- Bipolar
- Depression
- Normal
- Personality Disorder
- Stress
- Suicidal
Generates a personalized mental wellness report based on:
- User responses
- DistilBERT predictions
- Behavioral assessment
The report includes practical recommendations and wellness insights tailored to the user's responses.
Hugging Face Model
https://huggingface.co/anisha-14/Mentify-DistilBERT
The complete notebook used to train and fine-tune the DistilBERT model is available here:
https://colab.research.google.com/drive/1ds8LwlH5gBoMOYptxi-l2LJUfT9z30Jf?usp=sharing
👨💻 Authors
Anisha Rawat
Aryan Singh
⭐ If you found this project helpful, consider giving this repository a Star! Your support helps us continue improving MentiFy.