Standardizing Fair, High-Quality Grading with Pensive

## Overview Instructor: Dr. Yi Zhang, Senior Lecturer, Department of Industrial Engineering and Operations Research, Columbia University Yi Zhang teaches data analytics, probability, statistics, and simulation courses at Columbia Engineering, typically serving medium-to-large classes of around 100 students per section. In Fall 2025, Yi taught two sections of Probability, Statistics, and Simulation with a combined enrollment of roughly 250 students—courses that are highly computational and require extensive coding-based assessment. Facing an increasingly heavy grading burden and concerns around grading consistency, Yi adopted Pensive during the Fall 2025 semester. His primary focus was on using Pensive’s AI-powered rubric creation and grading workflows to improve fairness, consistency, and feedback quality while **dramatically reducing time** spent grading. ## Challenge Before Pensive, grading posed a significant challenge for Yi and his teaching team: - Coding-based assignments and exams required manual review of lengthy student-created codes, mostly in Jupyter Notebooks. - Eight course assistants (CAs) spent large amounts of time grading, leaving less time for office hours and student-facing support. - Rubrics were time-consuming to design and deploy across multiple graders - Providing detailed, personalized written feedback at scale was difficult, especially under time constraints. > As instructors, we care deeply about fairness and consistency. But without strong rubrics and enough time, it’s very hard to guarantee that every student is graded by the same standard. > > — Yi Zhang Yi had already begun experimenting with AI tools on his own—copying student code into ChatGPT and manually guiding it with grading criteria—but this ad hoc approach **lacked structure, consistency, and scalability**. ## Solution Yi adopted Pensive as a centralized grading platform, with a particular emphasis on AI-generated rubrics and AI-assisted feedback. His workflow quickly evolved into a structured, iterative process: 1. Upload assignments into Pensive 2. Generate an initial rubric using AI 3. Review and refine the rubric to align tightly with learning objectives 4. Use the finalized rubric to guide AI-assisted grading and feedback 5. Apply instructor judgment to adjust scores where needed > The rubric generation gave us a strong starting point. It forced us to think carefully, before grading even began, about what really mattered in each question. > > — Yi Zhang Rather than replacing human judgment, Pensive became a **partner in the grading process**, helping Yi and his team standardize expectations and apply them consistently. ### Rubric Creation was a Game Changer For Yi, the greatest impact of Pensive was not speed, but consistency and clarity through better rubrics. - Pensive’s AI-generated rubrics helped define clear grading criteria across complex coding problems. - Yi refined the AI-generated rubrics to reflect precise learning objectives, ensuring alignment between instruction and assessment. - All graders, both human and AI, worked from the same structured framework, dramatically reducing variation in scoring. > Previously, grading could vary from one submission to another, even if unintentionally. Having strong rubrics in place ensures fairness and gives students confidence that they’re being evaluated consistently. > > — Yi Zhang Yi also found that the rubrics **improved the grader experience**. CAs reported greater confidence in their grading decisions and appreciated the clarity provided by well-defined criteria. ### AI-Generated Feedback Beyond rubrics, Yi was particularly impressed by the quality of Pensive’s AI-generated comments. > The AI feedback is extremely detailed and personalized—often better than what our CAs would have time to write. > > — Yi Zhang When grading a midterm exam on his own, Yi reduced grading time from multiple days to just five or six hours, while still delivering thorough feedback to students. ## Impact on Teaching and Learning By reducing grading overhead and improving rubric quality, Pensive created additional benefits across Yi’s courses: - More time for instruction: Yi reinvested saved time into improving lecture slides, refining learning objectives, and designing more engaging class activities. - More student support: With grading taking less time, CAs were able to hold additional office hours and provide more individualized help. - Improved student confidence: Students reported feeling more confident about exams and course content, supported by clearer expectations and better feedback. Yi also observed improvements in teaching evaluations, which he attributes in part to clearer grading standards and faster feedback cycles. ## Comparison to Alternatives Yi had considered other grading platforms, such as Gradescope, but found them insufficient for his needs. > Traditional tools help organize grading, but they don’t support the kind of AI-driven rubric creation and feedback that coding-based courses really require. For data science and programming-heavy classes, that makes a huge difference. > > — Yi Zhang Pensive stood out as the only platform that aligned with his instructional goals while meaningfully improving grading efficiency and consistency. ## Conclusion For Yi Zhang, Pensive has transformed grading from a bottleneck into a structured, reflective part of the teaching process. By centering grading around high-quality, AI-assisted rubric creation, Yi achieved more consistent, fair, and transparent assessment—while reclaiming significant time for teaching and student engagement. > AI should be a partner, not a replacement. Pensive helps us grade better, not just faster. It gives us the structure we need to deliver fair, high-quality feedback at scale. > > — Yi Zhang As Yi expands Pensive usage to additional courses, he sees rubric-driven, AI-assisted grading as a foundation for more thoughtful assessment, better learning outcomes, and more thoughtful interactions for both students and instructors.