AskNow: An LLM-powered Interactive System for Real-Time Question Answering in Large-Scale Classrooms

Paper Title

AskNow: An LLM-powered Interactive System for Real-Time Question Answering in Large-Scale Classrooms

Publication Info

• Topic area: Real-time AI-powered educational support systems in large-scale classrooms.
• Keywords: Large language models, real-time learning, classroom AI, question answering, student engagement, instructor feedback, large-scale lectures, educational technology, speech-to-text, context-aware systems.

Background and Problem

• Problem / challenge: Large-scale classrooms hinder meaningful student-instructor interaction due to social anxiety, fast-paced lectures, and limited opportunities for real-time clarification. Existing tools like Audience Response Systems and backchannels provide limited support for personalized, immediate feedback.
• Significance: Addressing these challenges can improve student engagement, reduce confusion, and enhance learning equity in large-scale educational settings.
• Motivation and related work: Prior systems like SPHERE and Jill Watson demonstrate the potential of LLMs for educational support but focus on asynchronous or specialized contexts. Real-time AI systems for synchronous, large-scale classrooms remain underexplored, particularly in addressing both student and instructor needs.

Solution

• Proposed approach: AskNow, an interactive system powered by LLMs, provides real-time, context-aware question answering for students and aggregated confusion insights for instructors during live lectures.
• Novelty:
  1. Real-time transcription and context-aware question answering directly linked to lecture content.
  2. Anonymous, low-friction question submission to reduce social anxiety.
  3. Semantic clustering of student questions for instructor insights into collective confusion.
  4. Integration of student and instructor interfaces for synchronized classroom support.
• Procedure and key techniques:
  • Real-time speech-to-text transcription provides lecture context.
  • Students submit anonymous questions via a GPT-like interface, receiving immediate, contextually relevant answers.
  • Questions are clustered using K-means algorithms based on semantic embeddings and displayed to instructors in aggregated formats (e.g., by topic, top five questions).
  • Instructor interface includes tools for monitoring confusion areas and reviewing post-lecture insights.

Results

• Concrete findings:
  • Students’ perceived time to resolve confusion decreased significantly across three courses (mean reduction: 1.1–1.4 points on a 7-point scale).
  • Instructors rated AskNow’s responses highly for correctness (4.53–4.88) and satisfaction (4.45–4.75) on a 5-point scale.
  • Students rated usability dimensions positively, with "Ease of Learning" exceeding 6 out of 7 across courses.
• Advantage over baselines: Unlike clickers and backchannels, AskNow integrates real-time answers with aggregated confusion insights, enabling both immediate student support and actionable instructor feedback.
• Experiments / evaluation:
  • Deployment in three computer science courses with 117 students and 3 instructors.
  • Pre- and post-surveys measured perceived learning dimensions; instructors evaluated 300 question-answer pairs.
  • Semi-structured interviews with 24 students and 3 instructors provided qualitative insights.
• Limitations and future work:
  • Short deployment duration; limited to computer science courses at one university.
  • No objective learning metrics like quiz scores or retention tests.
  • Challenges with transcription accuracy and context length limitations.
  • Future work includes integrating multimodal materials, addressing privacy concerns, and exploring longitudinal impacts on learning outcomes.

Summary

AskNow leverages LLMs to provide real-time, context-aware question answering for students and aggregated confusion insights for instructors in large-scale classrooms. It significantly reduced students’ perceived time to resolve confusion and was rated highly for accuracy and usability. The system addresses barriers like social anxiety and lecture flow disruption while offering instructors actionable feedback on student misunderstandings. Future research should explore broader deployments, integrate richer course materials, and measure long-term learning outcomes to refine its educational impact.