Michael Mitsch

Document Verification Portal

This Streamlit app demonstrates an intelligent document processing (IDP) pipeline for insurance underwriting workflows. Users upload proof-of-address documents, and the app runs automated AI verification against a mock customer database record.

The portal includes real-time audit logging for each session, showing how document uploads flow through rules execution and AI determination. It showcases end-to-end IDP workflow design for a production-style underwriting use case.
live demo

API for AWS RDS in Python

This project uses Flask, Docker, AWS, and Python to create a database that can store any machine learning dataset in a MySQL database and interact with the data using a web API. I have created a MySQL database hosted by Amazon RDS with 3 tables whos structure is shown to the right. I have also created a Python web API using swagger. The web API can be run in a docker container.

Any dataset stored in a google sheet can be stored on my database by using the add data endpoint and giving it the google sheet ID to search for. Also, you can use another endpoint to store machine learning models that are trained using the datasets you load from the database. The MySQL database keeps track of datasets, data instances (linked to datasets by the GoogleSheetID foreign key), and machine learning models.
github repo for this project   youtube project demo

Cloud Machine Learning

The link leads to a project I worked on with Ethan Nguyen, Wyeth Michaelsen, and Drew Willis. We created a machine learning algorithm that would predict how much the beat saber community would approve of maps based on their meta data. The project structure is very modular and can work with many machine learning implementations. It can be run from a docker container and different web endpoints control the data aquisition, database storage, model training, and graphical results of model testing.

My contributions to the project were data aquisition (scraped data from online map metadata... we were in contact with the owner of the website I go data from so scraping was ok), dynamically gathering data from CSVs on google, creating our tree regression model, creating our graphs, and creating endpoints in our master.yaml that were connected to my work.

github repo for this project

Detecting Social Media Bots using Unsupervised Learning

This poster summarizes the research I did in the spring semester of my sophomore year. All programming was done in Python for this reseach. I joined this research project as a part of the UROC program at the Luddy School of Indiana University. My contribution to the project was a KMedoid model for predicting whether a twitter account is a human or bot to get a better understanding of clusters in the data that help prediction.

Reach out to me if you want more information on this project.

This Website

This one is a little silly to add because it only took about a day, but I did learn a lot about CSS, html, and using javaScript in a website. Shout out to w3schools.com for the html tutorial. Also, I wrote this website using Adobe Dreamweaver because Indiana University students can download it for free. I highly recommend Adobe Dreamweaver for any students wanting to create html from scratch.

github repo for this website

Cell Simulation Capstone

Worked with my senior capstone group to develop a tool to simulate cell interation in 3D space with Blender (animation software) and Python scripts.

Goo website

Goo github repo