SBIR Phase I: Introducing Riley: A Co-Designed AI Teacher Powered by Natural Language Processing for CTE Success

The broader/commercial impact of this Small Business Innovation Research (SBIR) Phase I project is to validate a proprietary artificial intelligence model designed to deliver adaptive instructional support for Career and Technical Education, defined here as applied coursework that prepares students for skilled careers. Millions of middle and high school students across the United States participate in these programs, yet instructional capacity has not kept pace with demand, contributing to reduced course availability and declining student engagement nationwide. These constraints affect learners in all regions and limit the talent pipeline for industries that rely on early technical preparation. This project centers on a novel, data-driven instructional model that advances scientific and technological understanding of how artificial intelligence can replicate core instructional functions, such as timely feedback, task guidance, and concept reinforcement, in practical learning settings. The technology is positioned at the intersection of artificial intelligence and immersive learning systems, with an initial market focus on secondary education programs seeking scalable instructional support. The value proposition lies in a durable, software-based model that improves learning continuity without proportional increases in staffing. Commercialization is anticipated through recurring institutional licensing. By year three, the technology is projected to impact tens of thousands of learners nationwide, with outcomes measured through engagement persistence, task progression, and course completion rates. This Small Business Innovation Research (SBIR) Phase I project investigates a layered artificial intelligence architecture for real-time, closed-loop instruction inside an embodied virtual reality (VR) engineering simulation, leveraging proprietary, access-controlled instructional interaction data without disclosing protected content. The central technical risk is whether tightly coupled competency-graph retrieval, Bayesian latent-state estimation, and supervised pedagogical policy learning can remain stable and instructionally valid under strict end-to-end latency and synchronization constraints imposed by immersive, spatially grounded tasks. The research objective is to demonstrate that a modular pipeline can infer evolving learner mastery from multimodal behavioral telemetry, select context-conditioned instructional actions, and generate curriculum-bounded language that is both technically correct and pedagogically aligned. The proposed work constructs a structured competency graph over Engineering I principles, trains an intent classifier on expert-annotated dialogue moves, and develops a lightweight Bayesian learner model calibrated to task performance traces. A constrained natural-language generation layer is integrated with a low-latency VR state manager to ensure bounded outputs and deterministic grounding. The prototype is evaluated for retrieval precision, policy robustness, timing jitter, and agreement with expert educator judgments. Anticipated results include evidence of coherent operation at immersive frame-time scales and a transferable foundation for expanding to additional competencies and pathways. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.