❌  Without CuraAdapt

• Every student sees the same lesson content

• Failed formative checks go unaddressed

• No personalization to interests or career goals; motivation wanes

• Gaps persist and compound across lessons

✓  With CuraAdapt

• Gaps detected automatically from in-lesson formatives

• Personalized reteach generated instantly, inline

• Anchored to each learner's field, interests, and background

• Mastery-triggered — only when and where needed

① PROFILE

Learner Profile Build

The student's interests, field of study, career goals, and topic background are compiled into a structured learner profile.

② DETECT

Gap Detection

CuraAdapt uses an internal knowledge graph of concepts in the course to detect gaps in student learning (mastery assessment).

③ RETEACH

Personalized Content Generation

CuraAdapt inserts inline targeted reteach material,  anchored to the student's own interests and goals, for each gap concept.

RQ1  ·  Learning

Does personalized reteach improve learning?

Within each learner, do gap concepts that receive CuraAdapt's reteach gain more on a neutral cumulative assessment than a matched held-out control concept?

RQ2  ·  Generation Quality

Can CuraAdapt reliably generate high-quality content?

Is the generated reteach instructionally accurate and genuinely personalized to the individual?

RQ3  ·  Acceptability

Do learners find it useful?

Do students rate the personalized reteach as more relevant, engaging, and helpful than standard presentation,  and prefer it?

Participants
N = 14 students in CSC 101 (AI Foundations) at URI. Diverse fields of study, CS experience levels, and first-generation status.

Materials
CuraCourse Chapter 2 (AI Foundations); concept knowledge graph; parallel cumulative Forms A/B; learner perception & preference survey.

Control Condition
One gap concept per participant is held out: standard lesson only, no personalized reteach added. Cancels practice, re-exposure, and regression-to-the-mean confounds within each learner.

Safety & Quality Gate
All generated reteach blocks are reviewed for correctness by the research team before any learner sees them. Post-study, all blocks rated by the full 5-member expert panel.

3.99 / 4

Expert Accuracy Rating

Mean instructional accuracy across all generated reteach blocks, rated by a 5-member expert panel,  with zero factual errors flagged across all 40 blocks.

100%

Genuine Personalization

Percentage of generated blocks rated ≥ 3/4 on personalization fidelity (mean 3.67), anchored to the individual learner's field, interests, and goals.

11 of 11

Preferred the Personalized Version

Students who expressed a preference chose CuraAdapt's reteach for the rest of the course over the standard lesson (exact p = 0.001); 71% said they would want the feature.

Expert Panel Agreement

Accuracy: 97% exact agreement · Fidelity: ICC(2,k) = 0.58

Reliable inter-rater agreement across the five-member heterogeneous reviewer panel (content author, college AI instructor, K-12 educator, CS-ed researcher, recent CS graduate). All reviewers rated all generated blocks — a census, not a sample.

Learner Preference

Among learners who expressed a preference, the personalized version was favored on every item (89–100%; primary item 11 of 11, p = 0.001).

On a head-to-head forced choice, learners favored the CuraAdapt reteach over the standard lesson; a separate before/after attitude survey showed no significant change, making the forced choice the more sensitive measure.