Feeling tired, frustrated, and worse than you felt before? Most patients experience this when consulting telehealth services for the best treatments. That's because providers today see patients as a mixture of billing codes rather than people who have simple, yet complex, questions about their health. Intelliquette addresses them with its question-based (rather than product-based) UI, which connects the right combination of symptoms with conditions that best reflect a patient's health. These questions, as well as their corresponding responses, are produced by our bespoke AI-powered LangChain Large Language Model (LLM) syndicate, forming a distinct combination of technical acumen from a GPT-2-Medium model. We also incorporate embeddings into our LLM-produced sentences using the SentenceTransformers library, which is responsible for semantic analysis. This pre-trains our model to best serve users who look for answers and a sense of trust from our application.
What became our biggest challenge wasn't the difficulty of producing an innovative product at a hackathon, but instead refining the viability of our service. Establishing a clear distinction between diagnosing patients, thereby making assumptions about their care, and treating our patients as well-intentioned people who happen to be unknowledgeable about their medical treatment provided us the confidence and assurance our project was both tangible and applicable to India's current telehealth market. This is not to say that several learning curves necessary for innovation led us to take countless detours throughout the past 48 hours: we switched across multiple back-end and LLM providers, which brought us to focus on the application's dependability. None of these challenges dismiss the potential of our tool, however. With additional fine-tuning to the LLM model and further resource allocation to our front-end development, subsequent versions of our application may include a chatbot to which users can express their concerns within their own semantics, allowing our predictive model to expand its semantic and sentiment analysis in addition to producing accurate medical treatment plans.
Tracks Applied (4)
Auth0
Postman
Postman
Discussion