AI Presentation Coach: University of Chicago

The Arizona State University Artificial Intelligence Cloud Innovation Center, powered by Amazon Web Services (AWS), collaborated with the UChicagoGRAD office at the University of Chicago to explore how AI can help students, faculty, and researchers become more effective presenters. The project focused on developing a proof of concept for an AI-powered video presentation analyzer that helps users practice presentations, tailor delivery to specific audience personas, and receive actionable feedback on communication style and delivery. The POC was designed to improve student presentation skills through self learning and demonstrate the feasibility of using AWS AI and machine learning services for real-time presentation analysis in an educational setting.

Students, faculty, and professionals often need to present the same ideas to very different audiences, from academic committees and grant reviewers to investors, employers, and public stakeholders. Strong presentation skills can directly influence communication effectiveness, credibility, funding opportunities, interview outcomes, and stakeholder alignment. Research from the International Coaching Federation shows that 72% of coaching clients report improved communication skills, highlighting the value of structured presentation feedback.

However, human coaching does not scale easily. A small number of coaches may need to support thousands of students, faculty, and researchers, and coaching is often limited by schedules, appointment availability, and business hours. Many presenters also prefer to practice first with AI feedback because it can feel more private, objective, and low-pressure than immediate human review.

This need is especially important for non-native English speakers, who benefit from repeated practice, real-time delivery guidance, and confidence-building support. UChicagoGRAD identified the opportunity for an AI-powered presentation coach that could provide persona-specific, 24/7 feedback on both what presenters say and how they say it, helping users prepare more effectively for high-stakes presentations.

The CIC team built AI Presentation Coach as a serverless, cloud-based presentation analysis tool on AWS with three main flows: Persona Setup, Presentation Practice, and Post-Session Feedback.

• AI/ML Analysis: We use AWS AI services to analyze the presenter’s video and audio stream. Amazon Rekognition Video provides real-time facial and motion analysis (e.g., detecting eye contact, head gaze, and expressions), while Amazon Transcribe handles live speech-to-text transcription. This combination lets us compute metrics like speaking pace, filler-word frequency, and vocal variety. These insights are fed back to the user almost instantly during practice.
• Session API & Streaming: The solution uses AWS Lambda and Amazon API Gateway to handle live presentation sessions. A web client streams the presenter’s webcam and microphone input into the AWS backend. AWS Kinesis Video Streams (or similar) securely ingests the live media, and Lambda functions coordinate calls to Rekognition and Transcribe for on-the-fly analysis. The results are then streamed back to the browser interface with sub-5-second latency, enabling a responsive, real-time coaching experience.
• Content Ingestion: Before a practice session, users can upload slide decks or documents (PDF, PPT, DOC) to the system. These files are stored in Amazon S3, and Amazon Textract extracts text and layout from the slides. Extracted content can be used to identify key topics and even generate potential audience questions. This ingestion pipeline ensures that the AI coach “knows” the presentation content and can tie feedback to specific slide topics or concepts.
• Frontend Experience: A lightweight web UI guides users through each flow. In the Persona Setup flow, users select or define an audience persona (with attributes like expertise level, priorities, and communication style). In Presentation Practice, the UI displays live metrics and visual cues (e.g., a speaking pace bar, real-time filler-word count, or a gaze heatmap) as the user presents. In future phases, a post-session dashboard will present personalized analytics and a PDF feedback report. The interface is designed to be simple and accessible (e.g., support for multiple languages) and aligns with UChicagoGRAD’s branding and usability needs.
• Admin & Analytics Loop: All session data and feedback signals are logged for review. After each practice session, the system records the video and metrics in a secure data store (e.g., Amazon S3 or DynamoDB). This lets coaches or administrators review performance and identify common issues. Collected analytics also help the CIC team iteratively improve the AI models and content: for example, highlighting which persona attributes need more refinement or where the AI might have missed a metric.
• Data Security, Privacy & Compliance: Because the solution may analyze student presentations, recordings, transcripts, slide content, and researcher materials, privacy and compliance are central to the design. The system is intended to support FERPA-aligned practices through student consent, role-based access controls, encryption in transit and at rest, and data handling aligned with University of Chicago retention policies. User data remains under UChicago’s control, is not shared or used for model training or service improvement, and researcher materials remain private and accessible only to authorized users.
• Infrastructure: The entire solution is defined with AWS Cloud Development Kit (CDK) for fully automated deployment. This includes serverless components like API Gateway, Lambda functions, S3 buckets, and integrations with Rekognition, Transcribe, and Textract. Using CDK ensures the stack is repeatable and maintainable, enabling rapid iteration on the POC and future production deployments

This solution helps address a common communication challenge in higher education and professional development: presenters often receive limited, inconsistent, or overly general feedback. By creating a structured and repeatable presentation practice experience, the University of Chicago POC supports more personalized coaching and better preparation for high-stakes presentations.

The solution is designed to help:

• Improve presentation effectiveness through persona-specific feedback
• Provide real-time coaching on delivery behaviors such as pace, eye contact, vocal variation, and filler words
• Give users a more intuitive way to practice and refine presentations over time
• Demonstrate how AI can support communication skill development in educational environments

Now more than ever, it is essential for graduate students to develop strong public speaking skills so they can communicate the societal impact of their work to a range of audiences - including the public, policymakers, employers, and investors. The AI Communication Coach is an innovative tool that will help train our students to convey their research more effectively and confidently. Working with CIC to create our AI Presentation Coach was a delightful experience. Not only did we end up with an inventive product, but we had fun collaborating with such an accomplished, creative team. With graduate students as the target audience for the AI Presentation Coach, it was a tremendous advantage to have students themselves helping develop the tool; they recognized, firsthand, how much the AI Presentation Coach could benefit their peers. 

- Brooke Carrell, Assistant Provost and Executive Director, UChicagoGRAD Experience at the University of Chicago.

While this POC is focused on UChicagoGRAD’s presentation coaching use case, the solution pattern has strong potential across higher education, workforce training, research institutions, and professional development programs. A persona-aware AI presentation coach could be adapted to support students preparing for thesis defenses, researchers pitching grant proposals, founders preparing investor presentations, and professionals refining executive communications. The modular persona framework and real-time feedback model also create a path for expanding to additional audience types and broader training scenarios. This expansion potential is consistent with the scope’s emphasis on a reusable persona framework and future path to production. Future roadmap capabilities could include multilingual presentation evaluation, helping users receive feedback on clarity, pacing, pronunciation, and audience alignment across multiple languages.