AI-Assisted Platform for Healthcare and Insurance

Project Overview 

Description: The project automates communication between an insurance web portal and hospitals, focusing on efficiently managing patient insurance payment codes and ensuring accurate information flow. The automation covers code entry, data validation, PDF generation, and error handling, enhancing overall communication and data integrity.

Project goal: The goal is to streamline the process of handling insurance payment codes by automating data entry, PDF generation, and validation while ensuring accurate communication between patients, hospitals, and the insurance system. The goal is to reduce manual intervention, minimize errors, and improve efficiency in managing patient insurance information.

How it works: The system automates the entry of patient codes into the insurance portal, generates and validates PDFs, and manages communication with hospitals and patients. It uses Selenium for web automation, Prefect2 for orchestration, and AWS for secure data handling and flow management.

Who can use: Healthcare providers, insurance companies, and administrative staff involved in managing patient insurance information and processing medical appointments. This solution is designed for entities seeking to automate and streamline insurance code management and data validation.

Use case: A hospital needs to efficiently process patient insurance codes and ensure accurate communication with the insurance company. This system automates code entry, validates PDF documents, and notifies both the hospital and the patient, reducing manual errors and improving operational efficiency.

Technical Challenges & Solutions

1. Validating Patient PDF Data

It was important to ensure the accuracy and integrity of patient data. The task was to validate the data in a patient’s PDF, sourced from an insurance web portal, against the data from the hospital’s records. Before encrypting and uploading these PDFs to S3, it was imperative to ensure that the data was consistent and accurate. Any discrepancies or mismatches would need to be flagged for review.

The Solution: Pattern Mechanism.

To address these challenges, a pattern mechanism was devised. This mechanism was built on the principle of flexibility and adaptability. Here’s how it worked:

– Pattern Dictionary: Each pattern was defined as a dictionary. Within this dictionary, keys were designated to look up specific data points in the PDF. For instance, a pattern might be defined to look for a patient’s name, date of birth, or a specific medical condition.

– Comparison with Hospital Data: Once the data was extracted from the PDF using the pattern, it was compared with the hospital’s records. Any discrepancies were flagged.

– Adaptability: The beauty of this mechanism was its adaptability. If there was a need to validate a new data point in the future, a new pattern dictionary was simply added. Similarly, if the structure of the PDF changed, the corresponding pattern dictionary was adjusted to match the new structure.

– Error Reporting: Errors were generated based on the mismatches identified during the validation process. These errors provided insights into the nature of the discrepancy, making it easier for the review team to address them.

Outcome and Benefits:

The patterning mechanism proved to be a robust solution to the challenges posed by the task. Some of the key benefits realized were: – Accuracy: The mechanism ensured that the data in the PDFs matched the hospital records, ensuring the integrity of the patient data. – Efficiency: The process was automated, reducing manual intervention and the associated errors – Flexibility: The system was future-proofed. Any changes in the PDF structure or new data points could be easily accommodated.

2. Errors handling via Screenshot System for Web Portal Interaction

We developed a webdriver for automating interactions with a web portal. Given the unpredictable behavior of the portal, especially during validation code processes, debugging was a challenge.

Challenges:

– Unpredictable Behavior: The web portal occasionally exhibited unexpected responses.

– Debugging: Identifying the root cause of errors during code validation was difficult without visual context.

– Data Security: Ensuring that error data, especially screenshots, remains secure.

The Solution: Screenshot System

– Reference Screenshots: We maintained a set of predefined correct screenshots representing expected portal behavior.

– Error Capture: If an unexpected behavior or error occurs, the system automatically captures a screenshot.

– Encryption & Storage: The error screenshot was encrypted and uploaded to S3 for security.

– Logging: A log message was generated for context along with the screenshot.

– Comparison: It is possible to easily download the error screenshot alongside the correct reference screenshot to pinpoint discrepancies for in-depth analysis.

Outcome: The screenshot system streamlined the debugging process, allowing us to visually identify and rectify issues. By comparing error screenshots with expected behavior, we enhanced the reliability of our webdriver interactions, all while ensuring data security through encryption and secure storage.

3. Webdriver Development Challenge for an Intricate Insurance Platform

We created a sophisticated webdriver to interface seamlessly with a complex insurance platform. This tool was designed to autonomously fill out various forms using patient data, subsequently downloading and rigorously validating the associated PDFs. Recognizing the potential bottlenecks of relying solely on UI interactions, we strategically shifted a significant portion of the logic to operate at the HTTP level. This approach not only enhanced the efficiency and speed of the operations but also ensured a more robust and high-performance outcome.

4. Encryption system, which allows us to upload files to s3 bucket encrypted

We developed a versatile encryption and decryption system using AWS KMS, capable of handling data of any size. This system facilitates the encryption of PDF files and screenshots, allowing secure uploads to S3.

5. Development of a Flow Management System with Fargate and Prefect2

Our organization required a comprehensive management system to oversee and control the various automated flows. This system needed to be accessible by multiple administrators, equipped with error notifications, and capable of executing retry mechanisms. Additionally, it was essential that certain flows, upon completion, could initiate other flows. From a financial perspective, we aimed for a cost-effective solution where expenses were incurred only during flow execution.

The Solution:

– Prefect2 for Management: We selected Prefect2 as our primary tool for the flow management system. It provided a robust platform that allowed different administrators to launch and monitor automated flows. Its built-in features facilitated error notifications, ensuring that any issues were promptly flagged and addressed.

– Integration with Fargate: We integrated our management system with AWS Fargate to achieve cost efficiency. This ensured we only incurred costs when a flow was actively running, optimizing our operational expenses. Fargate’s serverless compute engine was ideal for our needs, eliminating the need to provision or manage servers.

– Flow Triggers: Within our system, we implemented a logic that allowed certain flows to act as triggers for other flows. This ensured seamless operations and reduced manual interventions.

– Retry Mechanisms: Recognizing the potential for occasional hiccups, we incorporated retry logic. This ensured that if a flow encountered an issue, the system would attempt to execute it again, enhancing reliability. Outcome: By integrating Prefect2 with Fargate, we successfully developed a dynamic flow management system that was both cost-effective and efficient. This system not only streamlined our automated processes but also provided a reliable platform for administrators to monitor and control operations.

Technologies

Python, SQLalchemy, PostgreSQL, pandas, selenium, prefect2, pytest, sphinx, boto3, requests, alembic, pypdf, ChatGPT, GitHub Copilot, GitHub actions, CI\CD, Docker, Telerik ReportViewer. AWS services: ECR, ECS, VPC, RDS, S3, CloudWatch, IAM, KMS, Fargate, EC2. Insurance web portal: QaUS Hospital: SAP