Abstract:

Oral and poster presentations are high-stakes assessments across virtually every EdUHK programme, yet students receive sparse, subjective delivery feedback and have few opportunities for low-risk rehearsal. Existing GenAI tools (e.g., ChatGPT, Gemini, Canva AI) help draft slides or scripts but cannot analyze how a student presents, such as in voice, gaze, gestures patterns, or audience-facing engagement. PresentCoach-AI addresses this gap by integrating existing, mature AI tools, including OpenAI Whisper (speech-to-text and prosody), MediaPipe (pose, hand, gaze), open-source facial-engagement libraries, and an institutionally-licensed LLM (e.g., GPT-4-class), into a unified web-based coaching toolkit. Students upload or record a practice presentation (PPT talk, poster pitch, or capstone defence) and receive a triple-layer report: (i) Content analysis: slide/poster design, argument structure, rhetorical clarity (LLM); (ii) Speech analysis: pace, pauses, filler words, pitch variance (Whisper + prosodic toolkits, for English, Cantonese and Putonghua); (iii) Nonverbal analysis: eye contact, hand gestures, posture, facial expressiveness (MediaPipe + open-source models). An LLM coach then synthesises these signals into plain-language, rubric-grounded feedback and a revision plan. The toolkit will be piloted in two presentation-heavy courses over two semesters and evaluated with blind-rated rubric scores, system-measured delivery metrics, and student self-efficacy measures.

Code:

T0307

Principal Project Supervisors:

• FU, Yujun Eugene

Keywords provided by authors:

• Generative AI coaching
• Multimodal analytics
• Nonverbal communcation
• Presentation skills
• Personalized feedback

Start Date:

01 Sep 2026

End Date:

31 Aug 2027

Status:

Ongoing

Financial Year:

2025-26

Type:

TDG