Rubric-Based Assessment of Student Reports Using Small Language Models
Organizations: Faculty of Informatics, Masaryk University, Brno, Czech Republic
Abstract
Language models can analyze and assess student texts using predefined rubrics and deliver structured feedback, reducing instructors' workload. However, existing solutions often rely on external services that are resource-intensive, costly, and raise privacy concerns, making them unsuitable for constrained environments. This study evaluates an automated assessment pipeline with three key aims: design for low-resource environments, privacy, and focus on a novel context in cybersecurity education. The pipeline integrates lightweight language models for textual evaluation and a vision-language model for image interpretation. Responses from 105 students were collected over two semesters of an undergraduate course during a cybersecurity exercise on incident response. Results across different hardware configurations show that small models in low-resource environments can achieve up to 97% accuracy in text-only and up to 94% in rich-text assessments. This rivals or even outperforms a large model while adding the benefits of locally hosted assessment. Our solution can help educators and practitioners who need grading assistance without replacing human oversight, especially in environments where cloud-based solutions may be impractical. This study contributes an open-source, locally deployable AI-assisted scoring pipeline that runs on affordable hardware and can be extended to various types of student reports in computing education. As a result, it makes automated assessment accessible to a broad range of educators and institutions.
Figures & tables
| System | GPU | VRAM | CPU | RAM |
| System 1 (S1) | RTX 2080Ti | 12 GB | AMD EPYC 7F52 | 30 GB |
| System 2 (S2) | RTX 5070Ti | 12 GB | AMD Ryzen AI 9 XH 370 | 32 GB |
| System 3 (S3) | RTX PRO 2000 | 16 GB | Intel i7-11700 | 64 GB |
| System 4 (SR) | H100 NVL | 94 GB | 2 AMD EPYC 9454 | 1.5 TB |
| Criteria | incident _reports | phishing_site _disabled | security _advisory | user_contact _and_reset | ||||||||
| Model | S1 | S2 | S3 | S1 | S2 | S3 | S1 | S2 | S3 | S1 | S2 | S3 |
| gemma3 | 68.57 | 68.57 | 68.57 | 97.14 | 97.14 | 97.14 | 68.57 | 68.57 | 68.57 | 77.14 | 77.14 | 77.14 |
| llama3 | 68.57 | 68.57 | 68.57 | 91.43 | 91.43 | 88.57 | 85.71 | 85.71 | 85.71 | 85.71 | 82.71 | 80.00 |
| mistral | 68.57 | 71.43 | 71.43 | 85.71 | 85.71 | 88.57 | 77.14 | 82.86 | 80.00 | 85.71 | 85.71 | 88.57 |
| qwen3 | 71.43 | 68.57 | 71.43 | 97.14 | 94.29 | 88.57 | 65.71 | 65.71 | 71.43 | 74.29 | 77.14 | 80.00 |
| gptoss | 45.71 | 88.57 | 82.86 | 57.14 | ||||||||
| Criteria | phishing_email _address | phishing_email _picture | phishing_ website_picture | sections | ||||||||
| Model | S1 | S2 | S3 | S1 | S2 | S3 | S1 | S2 | S3 | S1 | S2 | S3 |
| gemma3 | 91.43 | 94.29 | 88.57 | 85.71 | 74.29 | 80.00 | 71.43 | 82.86 | 77.14 | 51.43 | 54.29 | 48.57 |
| llama3 | 82.86 | 77.14 | 74.29 | 82.86 | 85.71 | 88.57 | 71.43 | 77.14 | 77.14 | 31.43 | 31.43 | 31.43 |
| mistral | 80.00 | 85.71 | 85.71 | 82.86 | 54.29 | 65.71 | 74.29 | 80.00 | 74.29 | 31.43 | 28.57 | 31.43 |
| qwen3 | 88.57 | 80.00 | 82.86 | 60.00 | 60.00 | 68.57 | 68.57 | 71.43 | 74.29 | 43.86 | 34.29 | 34.29 |
| gptoss | 82.86 | 37.14 | 77.14 | 40.00 | ||||||||
Automated Grading
Automated grading uses machine learning and language models to score student work against answer keys or rubrics and provide feedback.
See all 95 papers