Curator: A Revolutionary Health Assistant

Inspiration

The idea for Curator stemmed from the need to offer more personalized, actionable mental and physical health insights. We recognized that many existing health applications provide general advice without addressing the underlying causes of health concerns. We wanted to create a tool that not only gives users tailored recommendations but also explains the reasons behind those recommendations, helping them better understand and manage their well-being.

What it does

Curator is an AI-powered health assistant that combines the power of causality and explainability to provide personalized mental and physical health support. It leverages the RAG (Retrieve and Generate) framework and Graph Neural Networks (GNN) for causal discovery, allowing users to uncover the root causes of their health challenges. By offering clear, data-driven explanations behind each recommendation, Curator empowers users to take informed actions to improve their well-being.

How we built it

We built Curator by integrating Langchain for seamless conversational interactions with users and RAG for retrieving relevant knowledge to generate personalized advice. We utilized Graph Neural Networks (GNN) to model and predict causal relationships between health factors. The system relies on user inputs to populate a causal graph, which is then analyzed to generate meaningful insights. The focus on explainability was central to ensuring users could trust and understand the advice provided.

Challenges we ran into

One of the primary challenges we faced was handling an accurate and relevant dataset. The complexity of mapping textual health data to meaningful insights required extensive fine-tuning to ensure accuracy. Additionally, managing large amounts of textual data and transforming it into actionable knowledge was difficult, particularly in handling the nuances of user inputs and health-related terminology.

Accomplishments that we're proud of

We are proud of achieving causal discovery within Curator, which allows users to uncover root causes of their health issues rather than just surface-level symptoms. Moreover, the explainability of Curator's recommendations is another significant accomplishment. By combining causal insights with user-friendly explanations, we created a system that fosters both trust and action.

What we learned

Throughout the development process, we learned how to better handle textual data and improve its processing to generate accurate, insightful recommendations. We also deepened our understanding of how to best serve users through RAG, ensuring the recommendations were not only personalized but also relevant to each user's unique health context.

What's next for Curator

The next step for Curator is to enhance its capabilities using attention-based models for more refined causal inference. By incorporating these models, we aim to improve the accuracy and depth of our causal predictions. Additionally, we are focused on fully deploying Curator to make it accessible to users everywhere, continuously improving and adapting based on real-world interactions and feedback.

Built With

• flask
• langchain
• python
• pytorch
• react