evidoria

← Back to browse

Good practice Imported

AI-Assisted OCR Grading Pilot for Vietnam's Restructured 2025 National High School Graduation Exam

Vietnam · Thái Nguyên · See the Vietnam profile

Evidence: Observational / pre–post Top 73% 40/100 · Ask Evidence Copilot about this practice

Researchers in Vietnam piloted an AI/OCR grading tool for the restructured national high-school exam, hitting 98.5% handwriting-recognition accuracy on 93 papers — an early, unscaled step toward automating grading for an exam now sat by 1.2m+ students.

98.5 %
Character-recognition accuracy (2025 pilot)
93 answer papers
Pilot sample size (2025 pilot)
4.0+ out of 5
User satisfaction rating among evaluators (2025 pilot)
AI-Assisted OCR Grading Pilot for Vietnam's Restructured 2025 National High School Graduation Exam

Details

Maturity
Pilot
Promoter
Thái Nguyên University of Education
Period
2025 pilot study; exam administered nationally each June (1.2m+ candidates in 2026)
Keywords
exam grading, OCR, national assessment, education measurement

Context

Following Vietnam's 2025 restructuring of the National High School Graduation Exam to add multiple-choice, true/false and short-answer formats, researchers at Thai Nguyen University of Education developed an AI-assisted grading pipeline combining a direct-answer sheet format with Mathpix optical character recognition (OCR) to read students' handwritten short answers.

Results

Using a five-stage Borg & Gall development model (regulatory analysis, system development, expert validation, dual-phase trials and pilot testing), the team tested the tool on 93 real student answer papers, achieving 98.5% character-recognition accuracy and user-satisfaction ratings of 4.0+ out of 5 among evaluators. The study was published in the peer-reviewed European Journal of Education Studies (Vol 12, No 4, 2025).

Conclusions

The pilot is small and has not been adopted for live grading of the exam, which in 2026 was sat by more than 1.2 million students at over 2,500 test sites, so the 93-paper sample should not be read as validated at exam scale. Separately, national authorities have addressed AI-enabled exam cheating through police coordination and detection pilots rather than through this or any AI-based grading system.

Implementation
Indicative cost
Low (< €50k) — Small-scale academic pilot using off-the-shelf Mathpix OCR; no national procurement or infrastructure cost reported.
Time to results
Short (< 1 year) — Single 2025 pilot study cycle; no scheduled exam-scale deployment.
Staffing & skills
Research team at Thai Nguyen University of Education (Tran Quang Hieu, Nguyen Thi Phuong, Tran Thi Thu)

Conditions for success

  • Five-stage Borg & Gall development model (regulatory analysis, system development, expert validation, dual-phase trials, pilot testing)
  • Use of Mathpix OCR combined with a direct-answer sheet format for handwritten short answers

Common failure modes

  • Piloted on only 93 papers, far below the 1.2 million+ student scale of the actual national exam
  • Not yet adopted for live exam grading, so exam-scale reliability is unproven

Where it fits

Governance type
academic research pilot (university-led)
Scale
small pilot (93 papers)
Income level
lower-middle-income

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful