Learning and AI jobs share one workspace
Overview, teacher, student, and AI operations surfaces share project state, taking an exam job from creation and recognition through human review to student results.
Back to products[ EDUCATION PRODUCT ]
EduVision OS: cloud exit and AI grading
A regional education platform that brings learning assets, exam grading, image recognition, and infrastructure state into one operable control plane.
Deliver the learning workflow and runtime together
More than moving servers
GPU inference, message queue, object storage, backup sync, and school access paths all enter the migration scope so cost and reliability remain reviewable.
Separate public access from production integrations
The public workspace uses anonymized reference data and stays read-only; real school identities, data, and inference services require authorized integrations.
Six paths make the platform operable
Video, books, exam papers, and learning assets enter one resource catalog.
Recognition, scoring, low-confidence review, and result release form one loop.
Image-recognition jobs run on a local GPU node with visible runtime state.
Upload, recognition, grading, notification, and backup jobs are smoothed and retryable.
Media, paper images, and recognition results follow school and lifecycle policies.
Critical assets, results, and job logs are copied to a recoverable location.
[ MIGRATION EVIDENCE ]
Cost reduction needs a reviewable path
The annual cloud-cost baseline is about ¥744,000, while the post-exit run target is about ¥180,000. Billing, inventory, depreciation, bandwidth, maintenance, and spare capacity must reconcile the result.
ANNUAL RUN COST¥744,000¥180,000Post-exit target
Four steps from teacher job to student result
Create an exam job
The teacher configures class, paper, and grading rules.
Enter recognition queue
The job enters the message queue and runs YOLOv5 inference on a GPU node.
Review exceptions
Low-confidence results retain boxes, rationale, and revisions.
Release student results
Scores, wrong answers, and learning guidance return to the student surface.