Faculty and Medical Students Save 15 Hours per Week Using Ora’s Curriculum-Aligned AI Across 11,752 Lectures from 161 Schools.
Ora AI Research Team. Internal pipeline audit.
For every medical-school lecture, students sort through external study content (AnKing flashcards; UWorld / AMBOSS / USMLE-Rx vignettes; Boards & Beyond / Sketchy / Pathoma videos) to find the subset that covers what the lecture taught, a sub-task that takes ~55–65 minutes by hand (convergence band of bottom-up and top-down estimates; see Method). At ~15 new-content lectures per preclinical week (Halperin 20241), it compounds to ~14–16 hours/week per student, ~420–480 per year, ~840–960 across preclinical training, roughly twenty-two full-time work-weeks. Ora's pipeline does the same task in a median of 37 seconds per lecture, producing ~50 aligned study items, measured on 11,752 lectures from 161 medical schools, which is ~85% of US LCME-accredited MD programs4 and ~60% of US DO programs. Faculty do the parallel work at the syllabus and block level when building course schedules; the same pipeline serves that workflow.
per preclinical student
per lecture (current)
to date
(~85% US MD, ~60% US DO)
Last-quarter algorithm revisions roughly doubled per-lecture vignette density (median 7 → 14), raised video coverage (67% → 79%), and tightened the flashcard mix (median 47 → 31). Total per-lecture artifact size held roughly constant.
~97× faster per lecture. Manual convergence band 55–65 min (wider extreme-method envelope 30–120 min) triangulated from Halperin 2024 ecosystem time and Pan 2022 per-item experiments. Ora runtime is the measured median across the current-pipeline cohort (n = 1,286 jobs, mean 67 s, p90 139 s).
Time reclaimed at scale
Scaling the per-lecture hour to the published preclinical lecture load (Halperin 2024: 18.5 hr/wk in lectures across 102 US schools, ≈ 15 new-content lectures/week):
| Window (per student) | Central | Convergence band |
|---|---|---|
| Per lecture | ~60 min | 55 – 65 min |
| Per week (~15 lectures) | ~15 hr | 14 – 16 hr |
| Per academic year (~30 wk) | ~450 hr | 420 – 480 hr |
| Per preclinical career (2 yr) | ~900 hr | 840 – 960 hr |
Convergence band is where bottom-up arithmetic (~50 items/lecture × per-item curation) and top-down anchor (Halperin 2024 ecosystem time × 35–40% curation share) agree. Wider extreme-method envelope spans 30 min – 2 hr per lecture (8 – 30 hr/wk); the title uses the tight convergence band.
~55–65 min by hand becomes ~37 seconds with Ora's current pipeline, per lecture (~97× faster). At ~15 new-content lectures per preclinical week, that's ~14–16 hours/week per student, ~420–480 per year, ~840–960 across preclinical training, about twenty-two full-time work-weeks reclaimed per student. Faculty face the same per-lecture sourcing task at the syllabus and block level and use the same pipeline. Across 11,752 lectures from 161 medical schools processed so far, roughly 12,000 person-hours of manual sourcing labor substituted.
Method
- Inputs. Lecture documents (PDF, PowerPoint, Word) submitted through the Ora student app.
- Stages. Document parsing → semantic-embedding alignment against Ora's content graph → write links to per-modality tables (flashcards, vignettes, videos).
- Runtime. Median 37 s end-to-end on the current-pipeline cohort (n = 1,286 jobs; mean 67 s, p90 139 s).
- Cohort. 1,142 lectures whose first successful mapping job ran on the current production pipeline (the cleanest cut for current-pipeline output).
- Counts. Distinct linked items per lecture across the three modality tables; reported as medians plus at-least-one-item coverage.
- Reach. 161 schools and 11,752 lectures computed across the full mapping-job population.
- Bottom-up. ~50 items/lecture × per-item curation times anchored on Pan 20223 controlled experiments and AnKing/UWorld/B&B workflow inspection → central ~76 min/lecture.
- Top-down. Halperin 20241 ecosystem time (~39 hr/wk in Anki + third-party online + Q-banks) ÷ ~15 lectures/wk × 35–40% curation share → ~46–60 min/lecture.
- Convergence band. The two methods overlap at ~55–65 min/lecture, the basis for the title's ~14–16 hr/wk range. Wider extreme-method envelope: 30 min – 2 hr/lecture. No peer-reviewed study directly measures per-lecture curation time; this is an explicit triangulated estimate.
The ~1 hr/lecture manual-curation figure is a triangulated estimate, not a single peer-reviewed measurement. The student-side estimate is anchored in published medical-student data; the parallel faculty claim is structurally well-supported but not measured here. This brief reports artifact size (~50 items/lecture), not mapping correctness; a dedicated correctness brief (M-2) is in progress. The density cohort (n = 1,142) is narrower than the full reach population (n = 11,752) so density reflects the current production algorithm rather than a multi-version average.
References
- Halperin SJ, et al. Are Medical School Curricula Adapting With Their Students? J Med Educ Curric Dev. 2024;11:23821205241228455. doi:10.1177/23821205241228455
- Levy J, et al. Exploring Anki Usage Among First-Year Medical Students. J Med Educ Curric Dev. 2023;10:23821205231205389. doi:10.1177/23821205231205389
- Pan SC, et al. User-Generated Digital Flashcards Yield Better Learning Than Premade Flashcards. J Applied Research in Memory and Cognition. 2022;11(4):542–557. doi:10.1037/mac0000083
- Liaison Committee on Medical Education. Accredited US Medical Education Programs. lcme.org/directory