Auto-scoring · Live deployment
NSD-Auto-Scoring
한국어·영어·수학·과학·사회 자유응답 자동 채점 풀스택. K-12 라이브 운영중.
Korean / English / Math / Science / Social free-response auto-scoring stack — live in K-12.
자동 채점 + 자동 루브릭 생성 + 품질 모니터링까지 통합. K-12 라이브 트래픽 운영. 모델 채점 일치도(QWK 0.8471)가 정답(GT) 일치도 0.8323 을 +0.0149 상회 — 인간 평가자보다 일관된 채점.
Auto-scoring + auto-rubric + quality monitoring integrated. Live K-12 traffic. Model agreement (QWK 0.8471) exceeds ground-truth-rater agreement (0.8323) by +0.0149 — more consistent than human raters.
Mini-dashboard · indicators only
Verified signals. 추세·비교·상태만 — 절대 수치 일부는 공개 보류.
0.8471
Model QWK
+0.0149 vs ground-truth raters
1,136
Sample size
full enumeration · A.1 config
5 / 5
Subjects in production
Korean · English · Math · Science · Social
Quadratic-weighted kappa · model vs ground-truth raters
1,136 samples · A.1 config · Δ +0.0149 (model exceeds inter-rater agreement)
Subjects in production
Each subject powered by a dedicated NSD scoring component · K-12 live
Included NSD products
- NSD-Essay-V2 / V3
- NSD-Essay-Evaluation-Stage3
- NSD-K12-Essay-Scoring (live)
- NSD-MathScoringEngine-v2
- NSD-ScienceIntegrator
- NSD-SocialV3-Orchestrator
- NSD-RubricTemplating-v1
- NSD-CTS-Scoring
Live K-12 deployment · QWK measured on full 1,136-sample enumeration.