ClassPrep Autopilot

ClassPrep Autopilot is a multi-agent pipeline that turns a syllabus and a session topic into a verified, instructor-approved class-prep bundle — slide deck, student reference card, prep podcast script, and platform-ready quiz exports — in about two minutes. The problem: preparing one live class session takes an instructor roughly two hours of skilled but mechanical production work — building the deck, writing narration, making a reference card, recording a prep podcast, and hand-building quiz questions in two different platforms. That is exactly the work that keeps teachers from the parts of teaching only they can do. How it works: the instructor enters a syllabus excerpt and a session topic; eight stages run in sequence — Planner, Curator, Deck Builder, Reference Card, Podcast Script, NotebookLM Pack, Question Bank, and Verifier. It is a structured production pipeline, not a chatbot and not a "chat with your syllabus" RAG app. Nothing leaves the app until the instructor reviews the full bundle and clicks Approve: exports return 403 before approval, by design. After approval, Platform Agents export the question bank natively — Canvas (QTI, all question types native) and Kahoot (.xlsx, with any type conversions honestly reported in an instructor file). A human is the gate; the agents do the production. Proof — a real class, not a demo fixture: on Nov 1 the pipeline prepared GERMAN 3250 "Citizenship in the Age of Technology," Week 11 (Dataveillance), taught live Nov 2. Full bundle in ~1.8 minutes of machine time plus ~15 minutes of instructor review, against a ~2-hour manual baseline. Canvas export: all 8 questions native. Kahoot: 3 native, 4 converted with a conversion report, 1 essay instructor-only — nothing silently dropped. The public deployment completed its hosted demo run in ~75 seconds with the Verifier at 40 pass / 0 flags. Built during the hackathon (Oct 28–Nov 4, 2026): the web app and eight-stage pipeline, the Verifier stage, the human review/approval gate, the Canvas and Kahoot Platform Agents, and the public deployment. Pre-existing and NOT part of the entry: the instructor's own class-prep workflow and teaching specifications (slide-design rules, quiz presets, course materials) predate the hackathon — they are the know-how the pipeline implements; the code implementing them was written during the hackathon. Demo inputs are the instructor's own materials; no student data is used anywhere. Bonus category: Best Agentic Use-Case — a multi-agent pipeline with an independent verification stage and a human approval gate. Roadmap (not built): live posting to Canvas/Kahoot, a Quizlet Platform Agent, and a Results Agent that closes the loop after class.