I'm Carlos Rodriguez. I'm an AI systems researcher whose work centers on representational fidelity, the degree to which AI systems accurately encode, reflect, and serve the humans they claim to represent.

My research spans two registers. On the mechanistic side, I audit how language models internally encode demographic identity and whether those internal representations actually drive what the models say (Beyond Behavioral Benchmarks). On the systems side, I design coaching LLMs that practice restraint, paired with evaluation pipelines that scale past what human coding can support (Decision Coach).

That research didn't appear out of nowhere. I came up through computer science, then graduate design strategy at the Institute of Design at Illinois Tech, and spent years as a product and program manager across healthcare and education at Northwestern Medicine and Microsoft. The PM discipline still shapes how I work. I build the measurement instrument before I build features. I treat constraints as material, not blockers. I ship things real users touch.

Today I split my time between research, applied AI work in healthcare and education at Northwestern Medicine, and an Ambassador role at Anthropic supporting Illinois Tech's campus AI initiative.

What ties it together: I want AI systems that amplify human judgment instead of replacing it, and I want the evaluation instruments to be honest enough to tell us whether they actually do. If you work on alignment, evaluation, or applied AI in healthcare or education, let's talk.