Source-grounded assessments built through a six-agent workflow.
Recorded from the running local interface.
AQG Studio turns course materials into structured assessments. Its six-agent pipeline parses source documents, maps concepts, plans a balanced blueprint, generates grounded questions, evaluates them, and prepares reviewed exports. The workflow is built for educators and instructional designers who need traceable question sources and consistent difficulty settings.
The document processor extracts content and page or slide references from PDF, DOCX, PPTX, and TXT files. Knowledge analysis identifies topics and dependencies, then question planning allocates items across Bloom levels and difficulty tiers.
Generated questions are scored for groundedness, ambiguity, distractor plausibility, Bloom alignment, and bias. The refinement stage can revise weak items before the output agent creates LMS and document formats.
A Next.js studio, FastAPI services, LangGraph orchestration, document processing, retrieval-grounded generation, evaluation, and export paths.
Use separate workflow stages so source extraction, planning, generation, scoring, and export can be inspected independently.
Multi-stage generation takes longer and adds workflow complexity, but exposes the source, blueprint, quality checks, and revision decisions to reviewers.
A single assessment workflow covers source ingestion, question planning, grounded generation, quality review, and exports for common LMS and document formats.
The public GitHub repository is the canonical home for AQG Studio. You can read the commit log, fork it, file an issue, or send a PR.