Yu Huang, an assistant professor of Computer Science at Vanderbilt University’s College of Connected Computing, received the NSF CAREER Award in May 2026 for her research, “Enabling Human-Guided AI for Code.” The award is one of the National Science Foundation’s most prestigious recognitions for early-career faculty, granted to researchers with the potential to advance their fields and serve as academic role models.
Huang leads the MIND Lab — the Mixed INtelligence Development for Programming lab — where her team maps the cognitive processes underlying programming expertise to build more human-aligned AI systems. Using neurological imaging methods like fMRI and fNIRS alongside eye tracking and behavioral studies, they model the cognitive patterns behind code comprehension, error-making, and development of expertise. The goal is to feed those insights back into AI, building models that learn from human intelligence rather than working around it.
The Research
At the center of Huang’s CAREER work is a problem the field has largely overlooked: current AI tools for software development are trained on the artifacts that programmers produce — the code, the comments, the commits — but not on the cognitive processes that produce them. The mental effort, the strategic focus, the expert intuition behind every programming decision leave no trace in the training process. Huang’s research argues that these missing signals are precisely what AI models need to become genuinely effective.
“Humans are an extremely smart and complicated system that automated systems can learn so much from. But the challenge is how. That’s where my research lies.”
Her proposal addresses three interconnected challenges. First, feasibility: can the expertise of a skilled developer be captured and modeled? Second, scalability: how do you generate enough real human data to train AI at the scale those systems require? Third, integration: once you have the data, how do you build the training pipeline that makes the AI genuinely smarter from it? The result, she hopes, will be AI models that are cognitively aware — aligned with how humans think, not just what humans approve.
This is distinct from the familiar concept of “human in the loop,” where a person reviews or signs off on AI outputs. Huang’s approach trains the model on human cognitive patterns from the ground up, so the system reflects developer expertise rather than simply awaiting developer judgment.
A Broader View
Huang sees the rise of AI in software development as one of the most exciting moments in the field’s history. “AI is a great opportunity for us,” she said. “We are the ones who have the expertise to shape how AI gets built for software engineering, to make it smarter, more reliable, and more aligned with how developers actually work. That’s exactly what this research is about.”