Inspiration
- Strong Industry requirement for Conversational Chatbots in Medical Insurance domain which can understand caller's sentiments, context of questions and analyze both domain specific public knowledgebase and customer-specific private healthcare and insurance dataset.
- Another strong motivation is the availability of conversational AI/Gen-AI capability in Vertex AI. Now its extremely easy for the end users to build an Intelligent Virtual Agent.
What it does
- MIDAS is an intelligent assistant with deep understanding in medical insurance domain. It can also analyze customer information, claims, and policy details to answer questions and provide recommendations.
How we built it
- I used Vertex AI Agent Builder https://console.cloud.google.com/gen-app-builder/engines ## Challenges we ran into
- Few more iterations needed to verify the grounding and completion accuracy based on existing content. - Need to understand how to use the diagnostics for testing different
- Not sure how to deploy the app. The App using Javascript and html code generated using Dialog flow didn't work. It threw errors.
- Unable to make the DialogFlow part work as mentioned in instructions in https://cloud.google.com/dialogflow/vertex/docs/quick/create-application
Accomplishments that we're proud of
- Ability to generate very useful and meaningful conversations in medical insurance domain as demonstrated in the attached document.
What we learned
- how to create an Agent by providing relevant Goal, Instructions, Contexts and Constraints
- how to create tools and point to internal structured data and external knowledge base
What's next for MIDAS
- Add more Tools to enrich the Knowledgebase (indexed documents, code interpreters, site URLs etc.)
- Auto Generate Charts and Insights about healthcare and insurance info dynamically to further assist the internal users.
- Host the Chatbot in the Conversational Module of Sparkflows and show case to GCP Teams, Partners and Customers.
- Create a Business Friendly Analytical App for Medical Insurance Domain in our Self-Service Tool Sparkflows Fire-Insights using the API
Log in or sign up for Devpost to join the conversation.