Pull Requests From The Classroom: Co-Developing Curriculum And Code Conference: Mensch und Computer 2025; August 31-September 03, 2025; Chemnitz, Germany Mensch und Computer 2025 (MuC ’25), August 31-September 03, 2025, Chemnitz, Germany DOI: 10.1145/3743049.3748581 ISBN: 979-8-4007-1582-2/25/08 CCS
Dennis Zyska, Ilia Kuznetsov, Florian Müller, Iryna Gurevych · 2025
Read Pull Requests From The Classroom: Co-Developing Curriculum And Code Conference: Mensch und Computer 2025; August 31-September 03, 2025; Chemnitz, Germany Mensch und Computer 2025 (MuC ’25), August 31-September 03, 2025, Chemnitz, Germany DOI: 10.1145/3743049.3748581 ISBN: 979-8-4007-1582-2/25/08 CCS by Dennis Zyska, Ilia Kuznetsov, Florian Müller, Iryna Gurevych (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-00646.txt.
Introduction
Dennis Zyska and co-authors tell how a peer feedback tool for a writing course was built side by side with the course itself, in a paper from 2025. Course tools often miss what teachers aim to teach, which forces clumsy work rounds. Here the tool grew with the class, in step with each round of notes, swaps of drafts, and rewrites.
Joint growth tied tool traits tight to course goals, the team finds. Yet users hit snags in ease of use and in the tech base that ran the tool. Those snags taxed both staff and students. The authors argue that joint builds pay off only when teaching and tech teams plan as one crew from the start, with time set aside for fixes.
Catalog record
| CREATOR | Dennis Zyska, Ilia Kuznetsov, Florian Müller, Iryna Gurevych |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 24 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-00646
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Pull Requests From The Classroom: Co-Developing Curriculum And Code Conference: Mensch und Computer 2025; August 31-September 03, 2025; Chemnitz, Germany Mensch und Computer 2025 (MuC ’25), August 31-September 03, 2025, Chemnitz, Germany DOI: 10.1145/3743049.3748581 ISBN: 979-8-4007-1582-2/25/08 CCS: Human-centered computing HCI design and evaluation methods CCS: Human-centered computing Usability testing CCS: Human-centered computing Field studies CCS: Human-centered computing User studies Dennis Zyska, Ilia Kuznetsov, Florian Müller, Iryna Gurevych Abstract . Educational technologies often misalign with instructors’ pedagogical goals, forcing adaptations that compromise teaching efficacy. In this paper, we present a case study on the co-development of curriculum and technology in the context of a university course on scientific writing. Specifically, we examine how a custom-built peer feedback system was iteratively developed alongside the course to support annotation, feedback exchange, and revision. Results show that while co-development fostered stronger alignment between software features and course goals, it also exposed usability limitations and infrastructure-related frustrations, emphasizing the need for closer coordination between teaching and technical teams. 1. Introduction From assignment submissions to grading and communication, Learning Management Systems (LMS) such as Moodle or ILIAS have become a foundational infrastructure in higher education. These platforms are typically generic in design and functionality, offering a one-size-fits-all approach that often misaligns with the specific pedagogical goals of individual courses and educators ( Gamage et al. 2022 ; Cabero-Almenara et al. 2019 ) . As a result, educators feel forced to adapt their teaching to fit predefined workflows, or to repurpose existing features to approximate what they actually want to do ( Chugh et al. 2023a ) . Consequently, teaching and technology often feel misaligned for students and educators ( Luo et al. 2024 ) , creating inefficiencies, increasing administrative overhead, and limiting the positive impact of educational technology ( Akintayo et al. 2024 ) . To overcome these challenges, prior work called for closer integration of pedagogy and technology through tandem design of software and curriculum ( Akbar 2016 ; Eady and Lockyer 2013 ) . The promise of this approach lies in the idea that as pedagogical goals and software infrastructure evolve together, they can support coherent workflows, reduce friction, and preserve educators’ control over content delivery. However, we still lack insight into how such co-development plays out in practice, including the challenges and experiences of those involved. In this paper, we explore what this process looks like in practice. We present a case study of a large university course on scientific writing integrating a custom peer feedback system. This system enables students and educators to annotate and review each other’s submissions using structured tags and free-text comments. Each student reviewed two peers and received both peer and educator feedback, which they used to revise their submission. The platform was developed iteratively during the course, adapting to legal, pedagogical, and infrastructural needs as they emerged. The contribution of this paper is two-fold: First, we show how pedagogical, technical, and organizational factors interact in the co-development of curriculum and educational software. Second, based on mixed-methods data, we analyze how course and platform co-evolved, revealing both the potential for better pedagogical alignment and tensions around feedback, usability, and AI integration. 2. Background and Related Work Technology in Higher education Implementation research ( Century and Cassata 2016 ) studies the extent to which software implementation efforts achieve the specified goals. Implementation research has a long history in EdTech; for example, in their analysis of 46 empirical research studies with focus on technology implementation issues, Chugh et al. ( Chugh et al. 2023b ) highlight significant challenges such as technology and stakeholder barriers ( Ranbir 2024 ) or the ability of students and educators to cope with new technology ( Chugh et al. 2023b ) . While previous studies explored stakeholder experiences with existing implementations or established courses ( Munday 2016 ; Thomas G Ryan and Park 2012 ; Jin et al. 2022 ) , our study focuses on how software and curriculum are co-developed to meet pedagogical goals. Peer Feedback and Learning Peer feedback is essential in higher education, as it helps students critique peer work ( Topping 1998 ) and reflect on their own writing ( Cho and MacArthur 2010 ) , and supports learning at scale when providing direct instructor feedback is not feasible ( Søndergaard and and 2012 ) . Peer feedback is widely used both in university courses and in online education, and is an active research area in educational psychology ( Bauer et al. 2023 ) . To support peer feedback, digital tools like Juxtapeer ( Cambre et al. 2018 ) and PeerStudio ( Kulkarni et al. 2015 ) guide students through structured, rubric-based reviews. These systems are optimized for short-term, synchronous settings and support novel features such as comparing submissions side-by-side or routing them to active peers for quick turnaround. In contrast to these works, we explore a scenario of a semester-long university course with asynchronous participation, slower feedback cycles, and deeper curricular integration. CARE The C ollaborative A I-Assisted R eading E nvironment ( Zyska et al. 2023 ) was selected as the core platform for this course due to its unique capacity to support systematic data collection and workflow integration. Originally developed as a research tool, CARE’s architecture enables fine-grained text annotation with pre-defined semantic and color-coded tagging (i.e., Highlight, Strength, Weakness, Other) derived from Kuznetsov et. al. ( Kuznetsov et al. 2022 ) , and interaction logging. To operationalize CARE within a live university course, the platform had to be significantly extended, as the original scope did not fully address the pedagogical, procedural, and legal requirements of curricular integration. 3. Course and software development Goals, curriculum and stakeholders. Our study focuses on the Bachelor-level course “Introduction to Scientific Work” at the computer science department of a major European university, launched in Winter 2024. As part of the course, Students write a six-page exposé on a predefined computer science topic. Each student provides two anonymous peer reviews, receives feedback from peers and Educators , and revises the exposé accordingly. Developers are responsible for the development and deployment of the software. The Educators included one lecturer, two doctoral researchers, one postdoctoral researcher, and six teaching assistants. Developers were a team of five research assistants led by a doctoral researcher. In total, 193 bachelor Students enrolled in the course; 159 completed the course. Figure 1 . Touch-points of developers, educators, and students with CARE along the course. Color-coded regions indicate the stakeholder engagement with the system: red areas denote active contributions, green areas represent read-only or observational use, yellow marks phases where stakeholders share their experience or feedback. Timeline of developer, student, and educator engagement during an 18-week course, structured by stakeholder roles (developers, students, and educators), and color-coded by type of engagement: red for active contributions, green for observational use (read-only), and yellow for shared experience or feedback collection. Developers are actively involved in system preparation at the start, followed by continued support, last-minute integrations, and bug fixes during the course. They engage in observational data provision toward the end. Students begin with a five-week exposé draft writing phase, with no system interaction. This is followed by a four-week peer feedback period (active engagement), then a four-week exposé revision phase (read-only),