exam-paper-plugin

Composing an exam usually means hunting through Word and PDF files, picking questions by hand, adjusting the layout, and leaving room for writing—tedious and easy to repeat the same items. exam-paper-plugin collapses that into a single Agent prompt: upload a question file and it auto-detects multiple-choice, fill-in-the-blank, true/false, long-answer, and essay items, then generates a printable .docx and .html exam with configurable headers, two-column choice layout, and writing gaps. WMF/EMF formula images are even rendered as PNG previews in the HTML output.

The real differentiator is the built-in pure-Python smart question engine. Multi-view feature extraction, HDBSCAN clustering, FAISS vector indexing, and MMR diversity reranking all run locally, so you never pay LLM token costs just to skim through an entire question bank on every request. Cluster explanations support human labeling at the group level rather than item-by-item annotation, and a versioned evaluation suite tracks clustering quality, retrieval hit-rate, and system performance across parameter changes and data growth.

It suits teachers and curriculum teams who assemble exams repeatedly, as well as developers who want a reproducible, local question-retrieval step inside an LLM workflow. The base parser needs zero third-party dependencies, while the smart engine runs in an isolated .venv, so the two layers can be adopted independently without interfering with each other.