This single-author preprint presents an end-to-end self-hosted, API-free pipeline that converts lecture PDFs into multiple-choice questions (MCQs) using a local LLM plus deterministic quality control (QC). The final released artifacts are plain-text question banks with an explicit QC trace, and no LLM call is required at deployment time.
Key Points
Fully self-hosted / API-free workflow: lecture content is not sent to any external LLM service.
Converts lecture PDFs into MCQs using a local LLM plus deterministic QC.
A seed sweep over three dummy lectures produced 120 accepted candidates across 15 runs × 8 questions.
Hard QC checks cover JSON schema conformance, a single marked correct option, and numeric/constant equivalence tests.
A warning layer flagged 8 of 120 items for residual quality risks, making revision points explicit.
Releases a final 24-question set in JSONL/CSV format, suitable for Google Forms import.
Framed through the AI2L perspective, the workflow supports privacy, accountability, and Green AI in education.
Seine A. Shintani. Self-hosted Lecture-to-Quiz: Local LLM MCQ Generation with Deterministic Quality Control. arXiv (Preprint, Version 1, 2026).
DOI: 10.48550/arXiv.2603.08729
Keywords: Local LLM, MCQ Generation, Deterministic Quality Control, Self-hosted Pipeline, Education, AI to Learn (AI2L)