Thirty Years Later: My Non-Traditional Path to a Doctor of Engineering in AI

Research is rarely a straight line. For me, the path began thirty years ago when I first graduated with a Computer Science degree. Now, at age 50 and serving as a CISO in the healthcare sector, I'm officially a D.Eng. student at the University of Michigan Dearborn. I wanted to document my path as something that others can possibly learn from.

I've documented how I prepared over the last year across several roadmaps and summaries on this blog. For those considering a similar change of direction later in their career, or anyone looking for a structured way to enter the deep end of AI research this is a synthesis of how I got here.

The Foundation: From Syllabus to Specialization

My journey started with a humble 12-week syllabus generated by ChatGPT. I didn't just want to use AI tools; I wanted to understand the mechanics. That led me to Andrew Ng's Deep Learning Specialization on Coursera, which proved to be an excellent foundational resource.

About 8 weeks in, I realized that while the math was a distant memory, the intuition was accessible. Andrew's reassurance that you can grasp concepts without being a mathematical expert was encouraging. I always want to dive into implementation and see something work so I looked for a small project I could learn with. At the time AI Scribes were the topic of the day so I worked on a prototype Scribe that could record audio, convert to text using OpenAI Whisper and leverage open source models on my home server to translate the text to clinical notes. I used open source models rather than APIs to commercial LLMs with an eye to privacy thinking healthcare organizations would want data to stay local.

The more I progressed, the more excited I became about reading research papers and running my own experiments. I've been interested in AI and the ties to psychology and neurology my entire life but I discounted it in the early 2000s as science fiction and something I would never see come to fruition in my lifetime. Learning more about what is possible today and how computing has caught up to 20 years of research was very exciting! There's so much in this field that's ripe for progress and new ideas.

Discovering Reinforcement Learning

While Large Language Models are the current hot topic, I found myself drawn to Reinforcement Learning. Somewhere along the line in looking at all the various branches of machine learning I stumbled across RL. The functionality in today's LLMs is truly amazing and you can do great things with these tools but they're not the holy grail of Artificial General Intelligence.

Reinforcement learning appeals to me because it's somewhat modeled on how humans and animals learn through interaction. It can be run with far less computing resources and it aims to tackle AGI. Some of the world-leading research comes out of Richard Sutton's group at the University of Alberta. The Alberta Plan for AI Research lays out a roadmap to advance the field over the next 5-10 years, which lines up well with my doctoral intentions.

This brought me back to some of my interests when I was an undergrad and aspired to be a psychiatrist. The prospect of implementing animal learning functions in computers really resonates with me.

I committed to the University of Alberta's Reinforcement Learning Specialization on Coursera and started building implementations from scratch. My original roadmap was ambitious, maybe too ambitious. A month in, I had to update it after realizing I needed to focus more on the foundations and the implementation work I had planned was too ambitious to do in parallel to the learning.

Why a D.Eng. Instead of a PhD?

My first instinct was to pursue a traditional PhD. I quickly discovered that path wasn't straightforward for someone 20+ years out of school. Finding a supervisor willing to take on a non-traditional student with no recent academic record or research publications proved difficult.

I sent several emails to professors who were doing research in areas I'm interested in explaining my career, goals and asking for advice on finding an advisor and received very little, if any response. I tried attending conferences and introducing myself in person but I felt I wasn't really taken seriously and was quickly dismissed. One prof I spoke to actually made the slip of saying "maybe you should find someone intelligent to do the research", no joke. Couple that with the fact every program I looked at expected residency and full time attendance when I wasn't willing to quit my job to pursue a PhD and I wasn't off to a really encouraging start.

That's when I discovered Doctor of Engineering programs. First at Johns Hopkins and then the program at University of Michigan Dearborn. They offered a different model focused on people like me with years of industry experience to apply to research problems. The D.Eng focuses on applied research as opposed to purely theoretical and is tailored for working professionals. It would be nice to see these programs in Canada to encourage industry experts to contribute to academic research.

Preparing for the D.Eng.

It was around this time I got my notice of acceptance to the D.Eng program at University of Michigan Dearborn. Having completed the RL specialization, it became pretty apparent that I needed to do a deeper dive into the math. It's been 22 years since I graduated and I haven't been practicing math at all. It made a lot more sense to focus on that weakness and start a literature review than diving deeper into implementations.

My 2-month preparation plan before the January 2026 start date focused on two pillars:

1. Mathematical Rigor: I dedicated nearly 100 hours to refreshing Linear Algebra, Calculus, and Probability through DeepLearning.AI's Mathematics for Machine Learning specialization. The math in the papers I was reading wasn't "adding up" (har har) and I needed to fix that.

2. Literature Review: I established a structured workflow using Zotero and Git to work through Sutton's 10 foundational readings—from his 1988 paper on Temporal Differences to the 2023 Alberta Plan. This side project has evolved into something I call research-assistant where I keep all my notes, journal, todos, paper synthesis and experiment results in a git repo in markdown so I can use agents to help manage my research. I'd like to publish research-assistant eventually but it's a little rough right now and I haven't had time to clean it up or put any type of UI on it.

First Steps as a D.Eng. Student

With this self-learning and experimentation over the past year I feel really well prepared to start the D.Eng. Now that I've officially started, I've been eager to produce something research worthy. My method of learning has always been to understand things "under the hood". I don't feel I fully understand concepts until I learn them from the ground up.

To that end I've started working on a Python package called alberta-framework and using that to replicate results from foundational papers across the various steps of The Alberta Plan. My first replication—Sutton's 1992 IDBD paper—validated that the meta-learning approach works exactly as described 30+ years ago. There's something satisfying about confirming that a 1992 algorithm still works precisely as documented. For me that experiment runs with the alberta-framework JAX implementation in under 2 minutes on my home gaming PC. I can't imagine how long those experiments would have taken to run on 1992 hardware (and I'm too lazy to figure it out)!

Beyond replication, I've started exploring gaps in the foundational work. My early experiments on scale non-stationarity answer a question posed in The Alberta Plan asking whether IDBD and Autostep perform better on pre-normalized inputs or not. Spoiler alert, they do. I've collected real-world data from SSH honeypot logs from another side-project dubbed chronos-sec that exhibits dynamic scale changes that break IDBD even with normalization. Two weeks in and I'm already seeing how industry experience can surface research questions that pure academics might not encounter. It's early-stage work, but it's encouraging to see the preparation paying off.

Why Do This Now?

People often ask why I'm pursuing a D.Eng. at this stage of my career. In my 2025 review, I reflected on turning 50 and counting backwards from the end as I plan for the future. Middle age has changed my perspective considerably. Reading books like Flow: The Psychology of Optimal Experience and discovering Stoicism have really made me rethink what's important in life and where I want to use my remaining time. I'm not doing a D.Eng to get a degree or for career advancement. I'm doing it because I see it as a path to learning academic research rigour so I can spend the rest of my years contributing to AI research.

Safe applications of AI in healthcare truly has the potential to revolutionize how we deliver care to patients and ease the weight on our overburdened healthcare systems. Being a "non-traditional" student means I bring 25+ years of systems administration, software engineering and security leadership to the research. That's not something most PhD students walk in with.

I'm aligning my research with Step 12 of the Alberta Plan: Intelligence Amplification. The goal isn't to automate healthcare workers or IT staff out of their jobs, it's to amplify their decision-making capacity. RL agents that work alongside humans to make proactive suggestions, explain its reasoning, and know when to escalate to humans.

Advice for Late-Career Pivots

If you're thinking about AI research later in life, here's what I've learned:

• Build from scratch: Don't just import libraries; implement the math. You won't truly understand TD learning until you've written the update rule yourself.
• Find a framework: Align your work with a larger research vision. The Alberta Plan gave me a 50-year roadmap to contribute to, not just a standalone project.
• Document everything: This blog has kept me accountable. When you write about what you're learning, you're forced to actually understand it.
• Refresh the math: If it's been decades since your undergrad, invest the time. It's humbling but necessary.
• Leverage your experience: Your industry background isn't a weakness—it's a differentiator. Understanding Real-world problems and data matter.

The D.Eng. has started. I expect to complete by late 2028 or early 2029. Stay tuned for updates. Same bat time, same bat channel!