What the tool actually does
AISSA, short for AI-based Student Slides Analysis, is a web-oriented tool built to support formative feedback on oral presentation slides in higher education. Students upload their slides before presenting, and the system returns two distinct outputs: a numerical score for each rubric item, and qualitative feedback written in three parts, namely strengths, areas for improvement, and suggestions for improvement.
The distinction matters because these are not the same measurement. The numerical score is a rubric-based quantitative judgment tied to a five-point Likert scale with explicit descriptors. The qualitative text is generated language that explains and advises. A reader comparing tools should not treat a high score and encouraging prose as interchangeable evidence of slide quality.
The tool is organized into three modules: a Feedback Generation Module, a Management Module connecting the tool to SQL and MongoDB databases and the file system, and a Visualization Module. Feedback reaches students through interactive Learning Analytics dashboards rather than as a static report.
How slide features become a score
The feedback generation module extracts slide-level features such as text length, font type, number of lines, images, and other layout characteristics. These features, together with the slide content, are assembled into a prompt sent to ChatGPT 5.2.
Scoring is governed by a teacher-defined rubric. The default rubric, designed by oral presentation experts at UAM, contains nine criteria related to slide quality, including text readability, contrast, and visual balance. Each criterion is assessed on a five-point Likert scale with explicit descriptors. Teachers can adapt the rubric criteria through the dashboard, and those modifications are automatically incorporated into the LLM prompt used for feedback generation.
This is the mechanism the paper claims: the rubric, not the model alone, defines what counts as good. Because the rubric is editable and feeds directly into the prompt, two courses using AISSA with different rubrics are not producing the same measurement, even if both return five-point scores.
Validation and regeneration before display
The model returns a numerical score and a brief justification for each rubric item, plus the structured qualitative feedback. Before anything reaches a student, AISSA validates the generated output by checking constraints such as length and excluded terms, and automatically regenerates the feedback when necessary.
This validation step is a guardrail on form, not a check on correctness. It can catch output that is too long or contains banned terms; it does not establish that a score matches what a human expert would assign. Readers should keep that boundary in mind when interpreting the tool's reliability.
Dashboards for students and teachers
The dashboards were designed and implemented using the Plotly Dash framework. Each student receives an individual Student Dashboard where they can upload slides before an oral presentation and review previously submitted slides through a historical view. Once processing finishes, students can view and navigate their uploaded slides alongside the feedback, so scores and comments can be read against the specific elements of their presentation.
The Student Dashboard also exposes the evaluation rubric used by the tool, a set of teacher-authored tips for creating effective slides, and graphical visualizations of scores per rubric item. Students can additionally evaluate the perceived quality and coherence of the generated feedback, and that input is collected to inform future refinements of the tool and its feedback generation process.
The Teacher Dashboard lets teachers review the automatic evaluations for uploaded slides, including direct access to each student's individual dashboard. Teachers can complete the same evaluation rubric themselves and can edit and customize it for different presentation contexts and learning objectives. When a teacher-provided evaluation exists, students see those quantitative scores alongside the system-generated feedback, enabling a direct comparison between automated and human assessment.
The teacher dashboard also surfaces interaction logs capturing student engagement, aggregated into summary indicators such as how many times a student accessed the dashboard, reviewed generated feedback, or interacted with uploaded slides. A separate administrator dashboard allows creating, modifying, and removing students, teachers, and courses.
The usability number and its denominator
AISSA is currently being piloted at the authors' institution with approximately 150 undergraduate and master's students within the MOSAIC-F framework, which aims to enhance oral presentation skills through personalized and data-driven feedback.
The preliminary usability result comes from a narrower group: the System Usability Scale was administered to 30 students and yielded an average score of 83.38, which the authors describe as indicating excellent perceived usability. The denominator here is 30 respondents, not the roughly 150 students in the pilot, and the measure is perceived usability rather than grading accuracy or learning gain.
This is the paper's only reported quantitative outcome. It tells a reader that the students who answered found the tool usable; it does not tell a reader whether the automated scores are correct, fair, or aligned with expert judgment.
Why score agreement is still unmeasured
The paper is explicit that a systematic evaluation remains future work. The plan is to use presentation slides from the SOPHIAS dataset and compare the automated scores generated by AISSA with evaluations provided by expert professors from the authors' institution. That analysis is intended to examine the extent to which AISSA aligns with human assessment patterns, both in terms of statistical differences and absolute agreement, and to explore similarities and discrepancies at the criterion level between rubric dimensions.
Until that comparison is run, the tool's grading validity is an open question. The dashboards already support a side-by-side view of automated and teacher scores, so the infrastructure for comparison exists, but the paper reports no agreement statistic, no correlation, and no criterion-level accuracy result.
Additional planned work includes exploring the perceived usefulness, trust, and pedagogical value of the generated feedback; comparing different AI models, including commercial LLMs and local or open-source alternatives, to see which align most closely with human evaluations; and investigating multimodal data to improve feedback.
Transparency and data handling constraints
To support ethical and transparent use, students are informed in advance that their slides will be uploaded to an external GenAI server for automated analysis, and they are instructed not to include personal or sensitive information in submitted materials. The platform and feedback are available in English and Spanish, regardless of the presentation language.
These conditions shape what the reported usability result covers. The 83.38 score was collected from students who knew their slides left the institution's systems and who were told to exclude sensitive content. Any adoption decision inherits those same disclosure and data-handling requirements, and the language coverage is limited to English and Spanish.
What to watch when comparing automated and human scores
The most decision-relevant feature of this paper is not the usability score but the comparison the authors have set up and not yet completed. AISSA's default rubric has nine criteria scored on a five-point Likert scale, and the planned SOPHIAS evaluation will test alignment with expert professors both as statistical differences and as absolute agreement, with attention to criterion-level discrepancies.
That criterion-level framing is the right place to look. A tool can look acceptable in aggregate while diverging sharply on individual rubric dimensions such as contrast or visual balance, and the paper's own plan acknowledges this by targeting similarities and discrepancies between rubric dimensions rather than a single overall number.
For a reader deciding whether automated slide grading is trustworthy, the paper's own priority is therefore clear: the usability evidence exists at 30 respondents, the grading-validity evidence does not exist yet, and the criterion-level agreement study against expert professors is the result that would change the answer.
Sources used in this answer
AISSA Demo: AI-based Student Slides Analysis Tool for Automated Grading and Feedback
AISSA lets students upload presentation slides and receive rubric-based quantitative scores plus LLM-generated qualitative feedback through Learning Analytics dashboards; a pilot usability survey of 30 students yielded an average SUS score of 83.38, while agreement with expert professor evaluations remains planned future work.