Store the rubric in postmeta on each quiz_question record, with fields for expected_points (an array of concepts that earn marks), weighting (how many points each concept is worth), and reference_lesson_id (where the topic is covered). SleekAI reads these through mapped variables and instructs the model to score only against listed criteria.

Yes if the rubric is concrete. The system prompt instructs the model to enumerate which rubric points the answer covers and assign weighted scores, rather than producing a holistic opinion. Two similar answers should land within a point of each other. Logs let instructors audit any drift and tighten the rubric if needed.

Yes if you expose a write endpoint via Tool Use. The bot can update wp_learndash_user_quiz_attempts, the LifterLMS quiz attempts table, or whichever schema you use, plus drop a feedback note in postmeta. Most instructors start with the bot suggesting grades and a one-click approve flow, then move to auto-write once they trust the rubric.

Yes. The bot logs its rationale per question, including which rubric points it credited. Students can ask 'why didn't I get the point for X' and the bot quotes its reasoning. If the rationale is wrong, instructors override in the LMS admin and the next conversation reflects the corrected score. Conversation logs serve as an audit trail.

SleekAI doesn't run a plagiarism check, but the bot can flag answers that look templated or that don't engage with the specific scenario in the question. Combine with a separate plagiarism plugin if your course has high stakes. For lower-stakes formative grading, the bot's tone-aware grading usually catches obvious copy-pastes.

SleekAI is a one-time WordPress plugin license. You bring your own OpenAI, Anthropic, Google, or OpenRouter API key. A single graded answer costs a few cents at standard API rates, far less than the human grading cost per submission and well below the per-grade pricing of AI grading SaaS.

Yes. Modern frontier models grade short-answer prompts in dozens of languages. The rubric can be written in the same language as the quiz, and the bot replies to the student in the language they used. Multibot lets a Spanish-language grader and an English-language grader run on different course tracks.

The bot grades against the rubric you provide, not against its own opinion of the subject. As long as the rubric captures what 'correct' looks like, the model needs only enough domain knowledge to recognise the concepts. For highly specialised topics, link the rubric to a small set of reference lessons via reference_lesson_id and the bot grounds itself before grading.