mark-student-work-multi-agent-v2

Automated grading of student work from PDF submissions is handled by a hierarchical multi-agent system that maps attempts, grades responses in parallel, and applies taxonomy tags to produce a canonical JSON marking artifact and a Markdown learning report.

A multi-agent marking workflow for educational assessment is a coordinated process where specialized subagents handle structural mapping, parallel transcription, fast-pass grading, and deep-dive adjudication across sequential phases to assemble a complete marking artifact.

Multi-agent orchestration handles low-confidence grading items through a deep-dive adjudication phase, triggered after parallel fast-pass grading, to review uncertain responses and ensure accurate taxonomy tagging before final artifact assembly.

Yes, automated marking can process both student PDFs and optional answer sheets by spawning subagents to map attempts against answer keys during the initial phase before parallel grading and final report generation.

Outputs from a multi-agent student assessment workflow include a canonical JSON marking artifact containing graded responses and taxonomy tags, a derived Markdown learning report, and profiling telemetry measuring orchestration efficiency.

Multi-agent grading supports parallel processing by orchestrating subagents to grade responses in parallel chunks during the fast-pass phase, enabling efficient handling of large student work PDFs before deep-dive adjudication.