• 

  Old thumbnail

GPTMD

Inspiration

The inspiration for GPTMD came from a very real and frustrating experience: seeing the bills for a simple pair of eyeglasses. The cost felt unreasonable, especially for something so essential to daily life. That moment sparked a bigger realization — access to basic medical guidance is often expensive, slow, or locked behind barriers that many people can’t easily cross.

GPTMD was created to help lower that barrier by giving people a fast, accessible way to understand their symptoms and get informed guidance before deciding what to do next.

What it does

GPTMD is an AI-powered medical symptom triage chatbot.

Users can describe their symptoms in any language, and GPTMD:

• Understands the input regardless of language
• Asks intelligent follow-up questions to narrow down possibilities
• Stops once it reaches a confident, reasonable diagnosis
• Returns the diagnosis in the same language the user used
• Clearly states that the result is informational, not a replacement for professional care

The goal is not to replace doctors, but to give people clarity, direction, and peace of mind — especially when access to healthcare is limited or costly.

How we built it

GPTMD was built as a cloud-native application using Microsoft Azure:

• Frontend

  • Simple web-based chat interface
  • Real-time conversational flow

• Backend

  • Python + Flask API
  • Session-aware conversation handling
  • Hybrid diagnosis logic (rule-based + AI-driven)

• AI & Cloud

  • Azure AI Foundry / Azure OpenAI for multilingual understanding and reasoning
  • Carefully engineered prompts to ensure safe, structured medical triage
  • Azure Redis Cache for session persistence
  • Azure Web App for scalable deployment

• Deployment

  • GitHub-integrated CI/CD via Azure Deployment Center
  • Automatic redeployment on every code update

Challenges we ran into

• Designing the chatbot to ask the right follow-up questions instead of jumping to conclusions
  
• Managing conversation state across multiple messages and users
• Ensuring multilingual consistency, so users receive answers in their original language
• Preventing unsafe or overconfident medical outputs
• Integrating Azure OpenAI correctly with deployment, secrets, and environment variables
• Balancing rule-based medical logic with AI flexibility
• Making the system scalable without losing context or accuracy

Accomplishments that we're proud of

• Successfully leveraging Azure AI Foundry to power a multilingual, reasoning-based medical assistant
• Building a full end-to-end cloud application (frontend, backend, AI, deployment)
• Creating a system that adapts its questioning dynamically instead of using a static script
• Designing the chatbot to stop only when it reaches a meaningful confidence level
• Making healthcare guidance more accessible, especially for non-English speakers

What we learned

• How to architect and deploy AI-powered applications on Azure
• How to safely use large language models in sensitive domains like healthcare
• The importance of structured prompts and confidence thresholds
• How session management and state tracking dramatically improve AI conversations
• How cloud tools like Redis and Azure Web Apps enable real scalability
• How accessibility (language, cost, speed) matters just as much as technical accuracy

What's next for GPTMD

• Expand the medical knowledge base with more structured clinical rules
• Add confidence scoring visualization for users
• Improve UI/UX with richer chat features and accessibility support
• Integrate optional links to nearby clinics or telehealth resources
• Explore compliance pathways for regulated healthcare environments
• Add voice input/output for hands-free use

GPTMD is just the beginning — the long-term vision is to make high-quality medical guidance more affordable, accessible, and global.

Built With

• azure
• chatgpt
• microsoftaifoundry
• python