HealthPlanIQ: Insurance Chat Bot Project Overview

Project Description:

HealthPlanIQ is an interactive web application designed to provide users with personalized information and assistance related to health insurance. The project integrates a conversational chatbot powered by OpenAI models and a sophisticated SQL query generation system linked to a Snowflake database. The goal is to offer users a seamless experience when seeking health insurance information and to provide context-aware responses.

Disclaimer:

This bot is designed to provide insights about insurance plans. Before purchasing any plans, we strongly recommend to thoroughly read the documents and consulting subject matter experts.

Links

• ChatBot Application - link
• Codelab Document - link
• Video Demo - link
• LangChain - link
• Snowflake Tutorial - link
• Airflow - link
• Open AI Cookbook - link
• Streamlit - link

Architecture

Team Members

Aditya Kawale

• NUID: 002766716
• Email: [email protected]

Nidhi Singh

• NUID: 002925684
• Email: [email protected]

Uddhav Zambare

• NUID: 002199488
• Email: [email protected]

Few Snapshots

Application:

Airflow:

Data Sources

Dataset Details Source

Individual & families health plan**

Statewise issuers, health plans, premium rates for age/family types etc. link

Plan benefits (Benefits and Cost Sharing PUF) Healthcare plan benefits that consumers get after enrolling link

Health plan quality ratings (Quality PUF) Healthcare plan wise quality ratings link

Machine Readable PUF Drugs covered in the plans per issuer link

Application Flow

Gathering Data for the chatbot application

1. Airflow Pipelines

   • Pipeline 1 (load_insurance_plans.py): Captures the details regarding insurance plans like insurance issuers, geographic information, insurance plan name, premiums, deductibles, etc. from yearly uploaded EXCEL file
   • Pipeline 2 (load_insurance_plan_benefits.py): Captures details regarding benefits covered by different insurance plans from a yearly uploaded CSV file
   • Pipeline 3 (load_insurance_plan_covered_drugs.py): Captures details regarding prescription drugs covered by the insurance plans and their tier details. This data primarily comes from APIs hosted by individual insurance providers. A EXCEL file indicating the drug APIs helps in gaining access to issuer's API landscape
   • Pipeline 4 (load_insurance_plan_quality_ratings.py): Captures details regarding overall ratings, member experience ratings w.r.t insurance plans. This data comes from yearly uploaded EXCEL file

   The Airflow pipelines primarily scrape the web-hosted data by making HTTP calls, clean the data and stage the data in Google Cloud Storage buckets. These buckets later act as external stages for snowflake data syncing.

2. Snowflake Pipelines

   • Snowflake DW acts as the central repository for all the insurance plans related data.
   • Snowflake stored procedures are written for every table - they help in syncing the data from GCS CSV files to Snowflake tables using commands like COPY and MERGE
   • These stored procedures are parameterized, which expect a GCS saved CSV file, whose data needs to be synced with Snowflake tables

Initiate Chat

1. Input Information

   • Users start by entering their state, county, and optionally, the insurance plan ID on the home page.

2. Chatting with HealthPlanIQ

   • Users proceed to the HealthPlanIQ chat bot, where they can initiate conversations by typing questions or using provided suggestions.

Behind the Scenes

The magic happens behind the scenes through a multi-step process:

1. SQL Query Generator

   • User questions are sent to OpenAI via LangChain, utilizing an LLM model as an SQL generator.
   • LangChain creates a conversation chain with memory, engineering the LLM model with database tables and columns.
   • The model generates a SQL query to select relevant data based on user queries.

2. Query Validation and Data Retrieval

   • The generated SQL query is validated to ensure it is not a DDL or DML statement.
   • Valid queries trigger data retrieval from the Snowflake database.

3. Insurance Agent Analysis

   • The extracted data is sent to another OpenAI model, playing the role of an Insurance Agent.
   • The agent analyzes the data, providing answers to user queries and suggesting three follow-up questions.
   • The analysis is sent in JSON format using the StructuredOutputParser class from LangChain.

User Interaction

• The analyzed results are returned to the chat bot in JSON format.
• The chat bot displays the information to the user in a user-friendly format.

Steps to access HealthPlanIQ ChatBot:

You can be directly access HealthPlanIQ ChatBot from Streamlit Cloud via link

OR

1. Clone the Repository:

git clone https://github.com/your-username/your-repository.git
cd your-repository/frontend

2. Install Dependencies:

pip install -r requirements.txt

3. Create secrets.toml:

Create a secrets.toml file in the frontend/.streamlit directory with required secrets.

# secrets.toml

## Snowflake
user = "your_snowflake_user"
password = "your_snowflake_password"
account = "your_snowflake_account"
warehouse = "your_snowflake_warehouse"
database = "your_snowflake_database"
schema = "your_snowflake_schema"
role = "your_snowflake_role"
client_session_keep_alive = true

# Open_AI
openai_api_key = "your_openai_api_key"
openai_model = "your_openai_model"

4. Run Streamlit App:

streamlit run Home.py

5. Access the App:

Open a web browser and visit http://localhost:8501.

Note:

• Ensure Python is installed.
• Review terminal messages for troubleshooting.
• Adjust steps based on your project configuration.

Features

• User-Friendly Interface: Guides users through easy information input.
• Information Collection: Collects state, county, and optional plan ID via a well-designed form.
• Input Validation: Validates state, county, and plan ID; checks plan ID in Snowflake database.
• ChatBot Interaction: Employs OpenAI-powered chatbot for natural language conversations and insurance info.
• Dynamic Interaction: Varies responses based on user location and plan ID.
• SQL Query Generation: Generates SQL queries robustly using OpenAI models.
• Context-Based Response: Integrates SQL query responses, suggests follow-up insurance questions.
• User Memory: Tracks chat history and queries for enhanced context.
• Continuous Interaction: Users ask more questions, seek clarification seamlessly.

Contribution

• Aditya : 33%
• Nidhi : 33%
• Uddhav : 34%

Individual Distribution

Developer	Deliverables
Aditya	Airflow pipelines
Aditya	Snowflake data modeling
Aditya	Github code integration
Uddhav	Prompt engineering to retrive SQL query.
Uddhav	Q&A ChatBot model creation
Uddhav	Streamlit application and deployment
Nidhi	Data engineering and cleaning
Nidhi	Query generator to enchance efficency
Nidhi	Data Research and Documentation

WE ATTEST THAT WE HAVEN’T USED ANY OTHER STUDENTS’ WORK IN OUR ASSIGNMENT AND ABIDE BY THE POLICIES LISTED IN THE STUDENT HANDBOOK.