TheDreadedQuestion

This diagram illustrates the architecture of a medical chatbot system

Inspiration

We recognized the gap between symptom onset and access to reliable medical information. Many individuals turn to unverified online sources, leading to anxiety and misinformation. We set out to create an intelligent, AI-driven tool that provides users with clinically relevant insights based on their symptoms.

What it does

TheDreadedQuestion is an AI-powered chatbot that accepts free-text symptom input and suggests possible conditions. It uses a textual entailment model and a structured knowledge base to provide accurate, real-time responses in natural language.

How we built it

We developed a modular system using:

A fine-tuned textual entailment model (based on AllenNLP) for condition inference
Rasa for conversational management
A multi-agent knowledge base to retrieve condition-specific details
A custom NLI dataset with over 850K premise-hypothesis-label samples

Challenges we ran into

Key challenges included:

Generating medically realistic data at scale
Training the model on limited hardware over several days
Interpreting vague or incomplete symptom descriptions
Balancing clinical precision with conversational clarity

Accomplishments that we're proud of

We successfully implemented a domain-specific NLI model, integrated it into a responsive chatbot, and created a scalable system capable of understanding unstructured health inputs. The end result is a practical, user-centric diagnostic assistant.

What we learned

We deepened our understanding of textual entailment in clinical NLP, multi-agent system design, and real-time AI integration. We also learned the importance of domain adaptation and context-aware dialogue in healthcare applications.