Drupal AI Factchecker — Claim-Level Truth on AMD

Universities already pay fortunes for plagiarism detection — but plagiarism tools only answer "was this copied?", never "is this true?". AI-generated essays with fabricated facts sail through undetected. Drupal AI Factchecker is a claim-level content-integrity suite built as a native Drupal module (GPL-2.0+), not a bolt-on SaaS product. One click on any content node runs three checks: atomic claim extraction with per-claim verdicts (SUPPORTED / UNSUPPORTED / CONTRADICTED) verified first against the institution's own indexed corpus, an AI-writing likelihood score, and a verbatim plagiarism search. Every routing decision and model call is logged and auditable. All inference runs on AMD hardware: served with ROCm + vLLM on an AMD Instinct datacenter GPU during development, and the public live demo runs Gemma on ROCm on a Ryzen AI mini-PC — the same module and configuration scale from edge to datacenter. Student and research data never has to leave institution-controlled AMD hardware. Everything — models, evidence sources, trusted/distrusted domain reputations (seeded from Media Bias/Fact Check ratings) — is Drupal configuration, not code. It's built on the mature, pre-existing ai_provider_universal GPL module already on drupal.org; this hackathon added the AMD ROCm serving layer and the university content-integrity use case. Any university already running Drupal — a large share of .edu sites worldwide — can install it as a module, no new platform, no vendor lock-in, no black box.