Company Background:
The customer, Dong Zong, also known as the United Chinese School Committees’ Association of Malaysia, is a Malaysian education association serving the Chinese independent school community. Operating within the education and non-profit sector, Dong Zong supports examination administration and academic governance across a network of schools, where standardized, accurate, and auditable assessment processes are critical. As an education-sector organization managing high-volume examination workflows, the customer faced operational challenges from manual marking of shaded multiple-choice answer cards, including inconsistent grading, processing delays, and human review bottlenecks. Their adoption of an AWS-based AI-assisted marking and anomaly detection platform reflects a broader digital transformation effort to improve examination efficiency, reduce grading errors, strengthen auditability, and modernize paper-based academic operations.
Problem statement:
Dong Zong operates within a traditionally manual school examination workflow, where multiple-choice answer sheets were historically marked by human reviewers. This manual process created operational bottlenecks, inconsistent grading outcomes, and avoidable human errors, especially in cases where students partially erased answers, shaded multiple options unclearly, or made corrections that were difficult to interpret consistently.
Solution:
Webby delivered an AWS-based AI-powered marking platform to digitalize and automate the evaluation of shaded multiple-choice answer cards. The solution ingests scanned answer sheets, identifies student response patterns, interprets shaded selections, detects ambiguous or unusual markings, and flags exceptional cases for human review. This enables Dong Zong to move from manual paper-based marking toward a more scalable, standardized, and auditable digital marking workflow.
AWS plays a central role in the solution architecture. Amazon S3 is used for secure storage of scanned answer-sheet images, while backend orchestration can be implemented using Amazon API Gateway, AWS Lambda, and supporting observability/security services. For the AI layer, Amazon Bedrock can be used to evaluate managed multimodal/genAI capabilities, including Amazon Bedrock Data Automation for extracting structured insights from image inputs and Bedrock-hosted models such as Claude for exception analysis and ambiguity reasoning. AWS documentation confirms that Bedrock Data Automation is designed to transform images and other unstructured content into structured outputs, including through image-specific custom blueprints, while model modality support in Bedrock varies by model and should be explicitly evaluated during selection.
Outcome:
The business outcome of this transformation is improved marking consistency, reduced manual review workload, reduced grading errors, faster examination processing, and stronger auditability of examination results. To make this example fully competency-ready, Dong Zong’s submission should include quantified before/after metrics such as marking error-rate reduction, turnaround-time reduction, number of answer sheets processed, percentage of papers auto-marked without intervention, exception-flag rate, and operational hours saved.