AI Course Tutor Companion

AI Prompts to Build This

Copy these build prompts into Claude, Cursor, or your AI coding tool. Create a free account to unlock the full research behind them.

1. 1. Project Setup

   Build the weekend MVP of "Course Tutor Companion": upload a course's materials and get a student chat that answers only from them, with citations.
   
   Stack: Next.js (App Router), TypeScript, Tailwind, Supabase (Postgres with pgvector, Row Level Security, Storage, Auth with email magic link), Claude (Anthropic API) for the answers, and an embeddings API that returns 1536 dimensions. Deploy on Vercel.
   
   Tables (Row Level Security on, each creator reads only their own rows):
   - courses(id, user_id, title)
   - modules(id, course_id, title, order_index)
   - assets(id, module_id, type, storage_path) where type is one of pdf, transcript, slide
   - chunks(id, asset_id, content, embedding vector(1536), metadata jsonb)
   - students(id, course_id, email)
   - conversations(id, student_id, messages jsonb)
   
   Screens: /login, /course (upload and modules), /tutor/[courseId] (the student chat).
   Env vars (names only): NEXT_PUBLIC_SUPABASE_URL, NEXT_PUBLIC_SUPABASE_ANON_KEY, SUPABASE_SERVICE_ROLE_KEY (server only), ANTHROPIC_API_KEY, EMBEDDINGS_API_KEY.
   
   Do not build: billing or plans, Teachable or Thinkific connections, struggle detection and digests, adaptive paths, voice, email to students, analytics.
   Done when: npm run dev starts, you can sign in, and the six tables exist with pgvector turned on and Row Level Security on.

2. 2. Core Feature

   Build the one feature that proves the idea: a tutor that answers only from the creator's own content, with links to the source.
   1. Upload: split each PDF, transcript or slide text into chunks, embed them, and store them in chunks with the module and asset in metadata.
   2. POST /api/chat takes a course id, a student and a message. Embed the question, fetch the top 8 chunks for that course with similarity above 0.75, and call Claude.
   3. System prompt: "You are a tutor for {courseTitle}. Answer ONLY using the provided excerpts. If they are not enough, say so and suggest the relevant module. Prefer Socratic hints when hint_mode is on."
   4. Return the answer plus citations [{ moduleTitle, assetType, excerpt }] and render each citation as a link to that excerpt.
   5. Save the messages to conversations. Refuse medical, legal and financial advice outside the course.
   Rules: ship the citation links on day one. Creators trust the tutor when they can click through to the exact paragraph.
   Empty state: with no uploads, show the upload box and one sentence on what is supported.
   Done when: uploading a sample PDF lets you ask a question it covers and get a cited answer, and a question it does not cover gets a polite "not in this course" plus a module suggestion.

3. 3. Landing Page

   Build a one-page landing site for Course Tutor Companion, an AI tutor trained on a creator's own course.
   Hero: "Turn any online course into a 24/7 AI tutor trained on your content." One button: Join the waitlist.
   Sections: the problem (2am student questions and drop-off), how it works in four steps (upload, train on your content, 24/7 answers with citations, see where students get stuck), a sample answer with a citation link, and an FAQ on accuracy, what it will not answer and student privacy.
   Waitlist: store the email in a waitlist table in Supabase. No other service.
   Voice: warm, practical and respectful of the creator's expertise.
   Done when: the page renders on a phone and a submitted email appears in the waitlist table.

4. 4. Branding Package

   Use a design or image tool for this one. A coding agent cannot draw a logo.
   Brand for Course Tutor Companion: a wordmark and an icon that suggest a helpful margin note.
   Colors: warm off-white, deep ink and one friendly accent. Type: one readable sans for the interface and a serif for quoted excerpts.
   Deliverables: wordmark, icon, a citation chip design, and one launch graphic.
   Done when: each deliverable is saved in one folder and the citation chip reads at 14 px.