Sehrish Basir Nizamani

Research project

Classroom analytics

Feedback for instructors, privacy for students.

Status
Active
Since
2025
Themes
Educational technology, Human-centered AI

Overview

Knowing how engaged students are during a lecture has traditionally meant trained observers, recorded sessions, or instructor intuition, and all of those are slow, expensive, or raise privacy concerns. This project builds a different pipeline. It analyzes classroom video to extract engagement signals, then permanently deletes the footage before any results are saved.

What remains is geometric data only: skeletal coordinates and gaze vectors. A language model turns that data into engagement timelines, attention heatmaps, and behavioral summaries that appear on an instructor dashboard after each session. The aim is formative, reflective insight for the instructor, not surveillance or grading.

Research questions

  • Can an LLM reason about student attention from pose and gaze data alone, with no video or images?
  • Can engagement be analyzed in a way that preserves privacy by design?
  • What do instructors find useful in an automated, formative view of their own class?

Methods

  • Face blurring before any other processing, then OpenPose to extract 25 skeletal keypoints per person per frame
  • Gaze estimation with Gazelle, built on a frozen DINOv2 encoder
  • LLM analysis in 60-second segments, five-minute synthesis passes, and a final session summary
  • Original frames deleted immediately after extraction; only coordinate files are kept
  1. 2026
    ConferenceAccepted

    WIP: Beyond manual observation: Automated analysis of student engagement while preserving privacy

    Nolan Platt (student co-author), Sehrish Basir Nizamani, Yoonje Lee (student co-author), Alp Tural, Elif Tural, Saad Nizamani, Nada Basit, and Andrew Katz

    IEEE Frontiers in Education (FIE 2026), Work-in-Progress track

  2. 2026
    PreprintPreprint

    Can LLMs reason about attention? Towards zero-shot analysis of multimodal classroom behavior

    Nolan Platt (student co-author), Sehrish Basir Nizamani, Alp Tural, Elif Tural, Saad Nizamani, Andrew Katz, Yoonje Lee (student co-author), and Nada Basit

    arXiv preprint

    arXiv for Can LLMs reason about attention? Towards zero-shot analysis of multimodal classroom behavior

* Student co-author.

More research

Other projects

All research