The problem Harmony AI solves

Harmony AI addresses two major health challenges often overlooked in accessible healthcare: mental health support and skin disease detection.

Mental Health Support: With rising rates of stress, anxiety, and depression, access to mental health resources remains limited, costly, or socially stigmatized. Many people lack support during difficult moments, especially outside of traditional therapy hours. Harmony AI’s 24/7 chatbot provides emotional support, stress-relief techniques, and a safe space for users to express their feelings. By making mental health assistance readily available, it bridges the gap for those who need comfort and guidance without barriers.

Skin Disease Detection: Skin conditions, from mild allergies to more complex dermatological issues, are frequently misdiagnosed or go untreated due to lack of access to dermatologists or hesitation in seeking help. Harmony AI’s image-based AI model offers users a preliminary diagnosis of skin conditions, allowing for faster, more informed decisions on whether to pursue medical consultation. This feature saves users time and provides an accessible option for managing skin health effectively.

And many more problems like skin, air etc

Challenges I ran into

While building Harmony AI, we encountered a few significant challenges that tested both our technical and problem-solving skills:

Natural Language Processing (NLP) Accuracy for Mental Health Chatbot
One major hurdle was ensuring that the chatbot could accurately interpret users' emotional states and provide appropriate responses. Initially, the chatbot would occasionally misunderstand certain phrases or tones, which led to responses that were not fully supportive. To overcome this, we fine-tuned our NLP model using a diverse set of mental health dialogues, implementing sentiment analysis and training it to recognize a wider range of emotional cues. Regular testing and feedback from sample users also helped refine the chatbot's sensitivity and accuracy.

Skin Disease Detection Model Generalization
Another challenge was creating an AI model that could accurately detect skin conditions across diverse skin tones and conditions. Initially, our model performed inconsistently, as the dataset was limited in diversity. To address this, we sourced additional image data representing a wide range of skin tones and conditions, improving the model’s generalizability. We also implemented data augmentation techniques to enhance robustness, and, finally, the model started delivering more accurate and consistent results.

Optimizing Model Deployment for Mobile
Deploying both the NLP and image-recognition models on mobile posed a significant challenge due to computational constraints. We optimized the models by reducing their size with techniques like quantization and pruning, making them more efficient without compromising accuracy. This allowed the app to run smoothly on mobile devices, making Harmony AI accessible and user-friendly.

Each of these challenges helped strengthen our understanding of AI deployment and accessibility, pushing us to create a more inclusive and reliable solution for our users.