John Doremus

About Me

I'm a machine learning engineer at Knowtion Health, where I built a risk-scoring pipeline from scratch, an ML process that didn't exist at the company before I built it. Two production models, retrained weekly, generate three scores that now inform recovery decisions on millions of dollars in monthly insurance claim recovery.

I started at Knowtion Health as an Artiva developer, five years onboarding new clients onto the platform, extending and creating features, and building integrations into custom external programs that handled what Artiva alone couldn't do. That work is how I learned how the company's data actually fit together, and it's also how I ended up building the ML work in the first place: I was already inside the company and already well-informed of its data when the opportunity opened up.

Getting from there to the ML role took patience. My degree is in computer science, and the only formal machine learning I ever took was a single elective in college. I spent my own time afterward turning that class's theory into something I could actually pitch to the business, and kept making that case well before anyone acted on it. After more than a year, the opportunity finally came, and I built the whole pipeline: data collection, feature engineering, scoring and delivering results to the teams and platforms that used them, automated retraining, the works.

Outside of work, I've been slowly chipping away at Mandarin (HSK 1 so far), good practice at staying comfortable being bad at something new.

Career

1. Machine Learning Engineer

   2024 – Present

   Knowtion Health

   Day to day, I own two production models, each retrained weekly to guard against feature drift, together generating the three scores that feed into recovery decisions. Beyond the modeling itself, I built the model registration process that sets the bar for what counts as production-ready, combining ML and Operations criteria, and I write the model cards that document each model for governance under HIPAA Safe Harbor-compliant handling. I'm also currently building out SHAP-based interpretability, so the scores are explainable, not just accurate. The automation these models provide has added the effective capacity of several full-time reviewers.

2. Artiva Developer

   2019 – 2024

   Knowtion Health

   I onboarded new clients onto Artiva HCx, handling the configuration and implementation for each. I also built two of the systems that extended what the platform could do on its own: an integration service that pushed real-time claim status updates from Artiva out to an external API, and a preprocessing service that stripped duplicate content out of files before they went to our print vendor. The rest of my time went to ongoing maintenance and enhancements to the platform itself.

3. Systems Analyst

   2015 – 2019

   Convergys

   I maintained Artiva RM, the collections platform used daily by more than 200 agents across major clients, including Time Warner and eBay, writing platform logic in Artivascript along the way. One of the bigger wins was rewriting legacy mainframe call-logging and file-movement functionality as C#/.NET Windows services, which saved the company over $500,000 a year. I also led a core account number migration across more than 20 databases and updated platform security to meet PCI compliance requirements.

4. QA Analyst

   2014 – 2015

   Galmont Consulting

   I did manual QA testing, remotely, on a policy-holder registration website for Blue Cross Blue Shield, working within HIPAA compliance requirements the whole way. The work involved a lot of XML, and I wrote Python scripts on the side to speed up the testing process itself.