Evaluating Handwritten and Multimodal, Free-Style Responses in Algorithms and Data Structures: A RAG-LLM-Based Feedback Framework
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
This innovative practice full paper presents a retrieval-augmented large language model (RAG-LLM) framework for evaluating handwritten and multimodal, freeform student responses. In the age of AI, open-ended questions play a vital role in computer science and engineering education, aligning with the ICAP framework to promote deeper cognitive engagement. However, large enrollments pose significant challenges in assessing such responses and delivering highquality, personalized feedback at scale while minimizing attentional errors. To address this issue, we introduce a tool that leverages RAG-LLMs to enable scalable, automated assessment and feedback generation with interpretable reasoning. By incorporating domain-specific content, vector-based context retrieval, and evaluation validation, our approach aims to reduce hallucinations, errors, and biases in generative AI outputs—ultimately enhancing both feedback accuracy and instructional value. We applied the framework to 816 student submissions from a graduate-level Algorithms course (Fall 2023), focusing on responses identifying the Big O notation of recurrence relations, which were manually graded with curated feedback. Using five LLMs and three embedding models, we conducted prompt engineering and evaluated the pipeline across four quality metrics. Our results show that while LLM choice had minimal impact, the selection of sentence encodings significantly influenced evaluation outcomes. We also applied the pipeline to auto-assess 770 responses from the Spring 2025 offering of the same course, with positive and promising results based on student perception data.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.003 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it