A $750,000 National Science Foundation grant is funding the development of an AI conversational tutor at The University of Texas at Arlington that refuses to give students direct answers. Instead, the system guides learners through situational assignments by asking questions and offering hints, a design choice driven by concerns that students were outsourcing their learning to generative AI tools.
Shuchi Deb, associate professor in UTA's Department of Industrial, Manufacturing and Systems Engineering, leads the project alongside Mohammad Islam, associate professor in the Department of Computer Science and Engineering, and several departmental colleagues. The team is building a platform that combines two-dimensional and virtual reality interfaces to place students in workplace or industrial settings where they must collaborate and prioritize tasks.
How the AI tutor works in the classroom
In Deb's courses, instructors will teach skills and concepts through traditional methods, while homework assignments will integrate the AI tutor. The system provides instructions for each step and offers hints when students make errors. At the end of an assignment, it delivers a completed document with a performance analysis that students can use to revise their work - but only within the same platform, blocking them from copying prompts into a separate AI tool to extract answers.
"I saw my children and my students using AI, and I could see that they were not learning. They were just using AI to finish the assignment," Deb said. "Getting an answer and copying it into a document isn't learning and that concerned me as an engineer. If engineering students don't learn basic skills involved with engineering concepts, they are likely to make mistakes that could be costly or, in some cases, cost people their lives."
The conversational approach mirrors how a classroom teacher works alongside a student at their desk - interacting, prompting, and steering without handing over the solution. This method sits at the intersection of AI for Education and the underlying Generative AI and LLM technologies that power the tutor's dialogue capabilities.
Real-world testing with Fort Worth police
Deb has already begun testing the system outside academia. Working with the Fort Worth Police Department, she is helping develop training that uses real-life scenarios built around de-escalation principles. The AI assesses officer responses - detecting when an officer listens and shows kindness, when they pick up on a subject the person wants to discuss, and when they fail to connect. It then steers the interaction to give officers perspective and practice in a controlled environment.
The police training work is sponsored by the Fort Worth Police Department and funded by the U.S. Department of Justice under Grant No. 15PBJA-23-GG-06172-NTCP, with Sergeant Nathan Owens serving as project manager and Officer Jorge Lopez as instructor and training content designer.
Why this matters for IT, development, and education professionals
The project signals a shift in how institutions are responding to generative AI's presence in learning environments. Rather than banning the technology or accepting surface-level use, Deb's team is building guardrails that force engagement with the material. For developers and IT trainers designing internal learning systems, the approach offers a template: an AI that withholds answers but provides structured feedback, performance analysis, and revision pathways inside a closed environment.
Deb said the broader goal is to strengthen students' technical skills while building their capacity for ethical AI engagement. "AI has limitations, and we want students to understand how to use it, but they also need to gain an actual understanding of the skills and concepts that they will use in their careers," she said. For professionals building or procuring AI training tools, that dual focus - technical competence plus AI literacy - will likely shape procurement requirements and design decisions in the near term.
Your membership also unlocks:
