ABOUT CHRIS
I’m Chris DeWitt. I work full time as a forecast engineer and study applied data science part time at UNC-Chapel Hill, where I expect to complete my master’s degree in May 2027.
I’m also a proud graduate of UNC Charlotte. My time there changed my life and gave me the confidence to keep learning, building, and asking bigger questions.
I’m fascinated by the systems foundational to modern life. Sometimes that means markets, forecasting, and quantitative models. Sometimes it means machine learning, agent architectures, backend systems, game physics, procedural audio, or an idea that seemed much easier on Friday night than it did on Sunday afternoon.
After my current degree, I hope to continue graduate work in computer science, with an emphasis on machine learning and the software and computational systems that shape everyday life.
CURRENT RESEARCH
I’m developing a connected research program around reliable agentic systems and decision-making under uncertainty. The work combines software engineering, machine learning, quantitative methods, and cognitive science—and documents failures and limitations alongside successful results.
My current technical reports examine two foundational problems:
TR-2026-001 — Local Integrated Evidence-to-Scenario Workflow
A traceable agentic workflow that transforms synthetic macroeconomic evidence into a deterministic financial scenario while preserving provenance, authority boundaries, and an inspectable record of every decision.
TR-2026-002 — Evidence-Gated Model Selection
An evaluation harness that separates ranking model candidates from deciding whether the available evidence is strong enough to select one. Its current result is intentionally useful: no defensible winner.
The next stage of this research will investigate belief diffusion—how people and machine agents combine, transmit, and revise beliefs when making decisions from incomplete or conflicting information. This provides a bridge between my interests in finance, cognition, artificial intelligence, and complex systems.
These are working research artifacts, not finished claims. I publish the methods, assumptions, limitations, null results, and degraded runs because understanding why a system fails is often as valuable as watching it succeed.
LET’S COMPARE NOTES
I like meeting people who are learning in public, building just beyond their current abilities, or willing to share what they have learned.
If you are working on an educational game, an applied AI system, an evaluation harness, an agent, a skill, or something difficult to categorize, I would enjoy hearing about it. I am especially interested in meeting mentors, collaborators, and fellow tinkerers who care more about learning and building than looking impressive.