This repository contains example implementations of programmatic solutions I have developed and deployed in my professional career for stakeholders in the medical/clinical industry. My portfolio includes a mix of traditional automation methods and neural network powered "digital worker" automation. While I am happy to program anything required of me, I have so far focused on automated solutions because they save organizations time and resources.
I am interested in leveraging artificial intelligence combined with programmatic automation in a variety of ways ranging from practical business applications to unique scientific projects. For example, my latest natural language processing project shortened turnaround time and improved quality of email-related administrative tasks typically done by humans. Below is a list of projects that I have built and deployed to solve challenges for colleagues met in the course of my career.
GUI-based application that identifies inconsistencies in research project team contact information between a document in a shared drive and data in a content management website.
v1.0 capabilities are as followes:
- Categorize email (1 of 37 categories)
- Move email to Outlook folder (1 of 150+ folders)
- Move email to hard drive folder (1 of 150+ folders)
- Save email attachment to relevant location (1 of 150+ folders)
- Report on actions as they occur per email
Many tasks in research operations are better managed by artificial intelligence. To determine how many tasks an A.I. could handle, I developed two Deep Neural Networks (DNN) to manage specific management tasks of a shared clinops inbox. Both DNNs will be continually improved upon, with the ultimate goal being a single robust DNN that can remain operational 24/7, continually monitor emergent events in a research project, and react accordingly through a suite of programs and scripts that complete tasks normally completed by humans.
I once read that there were no easy methods to recursively travel through Outlook folders. To address this issue, I created two functions; outlook_depth_detector and outlook_depth_pathfinder. These functions can be modified to suit multiple types of recursive activities in Outlook.
| Time to complete task BEFORE script | Time to complete task AFTER script |
|---|---|
| 45 minutes | 25-35 seconds |
A regularly generated report required data points from five seperate reports, each in a different location. Any data point shared by two or more reports needed to match across the board. Originally, an employee would need to review all records in these reports on a weekly basis to find discrepancies, notify the team member assigned to the discrepant study of the the issue, request they address the issue, and track their progress to ensure they address the issue.
This script completed this task automatically, including emailing the appropriate team members with a formatted list of task and study number to address.
| Time to complete task BEFORE script | Time to complete task AFTER script |
|---|---|
| 7-10 minutes per document | 6-7 seconds per document |
We recieve updated Protocol documents from multiple studies at irregular times. These documents must be posted online to two different websites, placed in specific folders in a shared drive (while moving off their previous versions), and sending them to key stakeholders to update critical clinical documents. This must be done for each updated Protocol received.
This script handles all of the above described tasks in mere seconds.
| Time to complete task BEFORE script | Time to complete task AFTER script |
|---|---|
| 12-15 minutes per report | 10 seconds per report |
This script collects all matching reports and notifications for an adverse event, combines them into a formatted email for distribution, checks for previous emails regarding the particular event, and attaches them if the intended recipient has not responded yet.
| Time to complete task BEFORE script | Time to complete task AFTER script |
|---|---|
| 30-35 minutes per invite | 6 seconds per invite |
Example implementation to automate committee invitation package generation. Fills out templates, writes/formats emails, attaches completed templated documents to outlook email.
v2.0 update:
- Replaced most pandas-based data transformations with python native approaches.
- Wrapped everything within one class.
- Improved command-prompt instructions/interface.
Meeting Summary/Action Item Generator. Creates formatted outlook email with action items from a team meeting.
Miscellaneous Scripts. Miscellaneous scripts and programs made to help colleagues quickly complete various smaller tasks.
Document Migration Part II: document cleaner/uploader. Second step in migrating 1000 documents from an old clinical document repository to a new one. Previous anticipated time to completion was 4 weeks. This script completed the task in 2 hours.
Document Migration Part I: Batch document downloader/handler. First step in migrating 1000 documents from an old medical document repository to a new one. Previous anticipated time to completion was 1 week. This script completed the task in roughly 1 minute 30 seconds.
| Time to complete task BEFORE script | Time to complete task AFTER script |
|---|---|
| 15-40 minutes (depending on number of emails received) | < 60 seconds |
Files emails, saves attachments and emails to appropriate folder on hard drive and outlook.