AI
VeriMath: Human-in-the-Loop Generation and Verification of Curriculum-Aligned Mathematics Assessment Items
Publication: cs.AI updates on arXiv.org
Author: Quy P. Nguyen, Duong Q. Nguyen, Dinh P. T. Nguyen, Truong-Giang T. Le, Truong A. Bui, Vi H. Vo, Quang-Thinh Bui, My-Phuong Ngo, Trong N. Do, Thao P. Le, Phuc M. Nguyen, H. Nguyen-Xuan
Source published: 19/11/2026, 05:00:00 UTC
Briefing created: 19/11/2026, 05:05:34 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 19/11/2026, 05:03:24 UTC.
Briefings
VeriMath is a human-in-the-loop framework for generating curriculum-aligned mathematics assessment items, evaluated against the Vietnamese Grade 12 curriculum. Using 4,555 items and 50 generation requirements across two model backbones, the system produced verifier-accepted drafts for 94% of requirements. Expert evaluations confirmed high mathematical accuracy and alignment, with strict teacher-readiness assessments accepting up to 56.8% of generated items. Analysis of rejected drafts highlights persistent domain-agnostic challenges in formatting, statement validity, and solution-step completeness.
Key points
- VeriMath uses a human-in-the-loop framework to generate mathematics assessment drafts anchored to specific curriculum sections and cognitive reasoning levels.
- The framework achieved 94% acceptance rate for verifier-accepted drafts across two model backbones when evaluated against 50 distinct generation requirements.
Context
This research addresses the challenge of deploying generative AI in education by creating a rigorous, curriculum-constrained system. The study utilizes a national curriculum case study involving 4,555 items from the Vietnamese Grade 12 curriculum to demonstrate practical application and reliability in real-world educational settings.
Limitations
Evidence is limited to a single national curriculum case study and two model backbones. Results are based on a snapshot of the research at publication time, with no data on long-term performance, user adoption rates, or specific failure modes beyond domain-agnostic challenges.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 19/11/2026, 05:05:34 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeiboy6ddyb2wbuernnk6p6sxkkqkor4tnw7hqfdxk7dynxdfctzh7q