Improving Student Communication Skills with a Context- and Domain-Aware Agentic AI Approach

Effective oral communication is a foundational competency across professions, yet student skills and presentation confidence have noticeably declined due to pandemic-era remote learning and an overreliance on reading AI-generated scripts. While regular, low-stakes formative feedback is the established best practice for developing these skills, providing it systematically is prohibitively resource-intensive for instructors at scale. This project will develop a student-facing, domain-aware AI system that processes recordings of student presentations to deliver highly specific, multidimensional communication feedback. Developed over a 12-month period, the system will use multiple specialized agents to evaluate presentations, allowing students to iterate and improve their skills before formal instructor assessment.

• The Problem Addressed: In traditional under-resourced courses, students are typically assessed on communication through a single, high-stakes final presentation with no opportunities to practice and adapt. AI tools are frequently misused by students to write content rather than develop their personal delivery.  

• The AI/Technology Solution: Utilizing Microsoft Copilot Studio, the platform hosts an ecosystem of specialized feedback agents that evaluate a presentation along different dimensions such as concision, clarity, completeness, concreteness. The multimodal pipeline processes synchronized audio transcriptions and slide visuals.

• Instructor Customization: Instructors use an accessible pipeline to calibrate agents using their own domain-specific norms and exemplary student work, shifting AI from an automation shortcut to an iterative learning coach.