CourseGPT

CourseGPT is more than an AI app - it's a vision for a more responsive, personalized education system. By allowing professors to train Al models on-demand, students gain immediate access to course-specific materials and logistical details. Professors and TA's can sit back and relax with a personal AI assistant managing repetitive Q&A, while students can enjoy instant answers to questions for all of their courses in one place.

Live demo: https://course-gpt.herokuapp.com/ (wait 30s to boot up. If offline, run locally—see below).
Demo accounts:

• Student/admin demo (preloaded chats/analytics): grading@gmail.com / i_love_grading
• Professor demo (upload/train focus): prof.emily@gmail.com / profemilydemo

What it does

• Course-trained chat - ingest PDFs or pasted text per course, chunk, embed, and route student questions through an OpenAI model with relevant context.
• Multi-role - students chat and bookmark courses; instructors upload training text; admins see feedback and usage analytics.
• Chat UX - history, inline feedback, prompt suggestions, copy-to-clipboard, and fuzzy search across all conversations.
• Auth - email/password with JWT + cookie sessions, optional Google OAuth.
• Analytics - sentiment and feedback dashboards with charts and word clouds.
• Data seeding - the demo is pre-seeded with sample course website content so you can explore immediately.

Tech stack

• Frontend: React 18, Redux Toolkit, Chakra UI + styled-components, Chart.js/D3 visualizations, React Router.
• Backend: Node.js, Express, MongoDB/Mongoose, Passport (Google), JWT + cookie sessions.
• AI/IR: OpenAI (chat + embeddings), TensorFlow.js Universal Sentence Encoder, BM25 + TF-IDF utilities for similarity and keywording.
• Build/ops: Concurrent dev runner, Heroku deployment scripts, Nodemon for hot reload.

Project structure

/client      # React SPA
/server      # Express API, auth, embeddings, analytics
/docs        # Extra screenshots
/readme_images

Run it locally

1. Prereqs: Node 18+, MongoDB connection string, OpenAI API key (and org id if required).
2. Environment (server/.env):

   DB_CONNECTION_STRING=mongodb+srv://...
   JWT_SECRET=super-secret-key
   OPENAI_API_KEY=...
   OPENAI_ORG_ID=...               # optional depending on your account
   OPENAI_GPT_MODEL=gpt-4o-mini    # or gpt-4o / gpt-3.5-turbo
   EMBEDDING_MODEL=text-embedding-3-small
   TOKEN_LIMIT=6000                # optional cap for prompts
   GOOGLE_CLIENT_ID=...            # optional
   GOOGLE_CLIENT_SECRET=...        # optional
   

   For staging/production builds you can also set REACT_APP_ENV=staging when building the client.
3. Install dependencies (root script installs client + server):

   npm install

4. Start in development:

   npm run start:dev

   • API: http://localhost:3001
   • Client: http://localhost:3000 (proxy points to the API)
5. Prod-style run: npm run start (uses the server in prod mode; build the client first with npm run start:staging or the Heroku postbuild).

Using CourseGPT

1. Sign up or use the demo account.
2. Create or pick a course, then upload text (or paste) to train it; the backend chunks, embeds, and stores vectors in MongoDB.
3. Ask questions in a chat; CourseGPT embeds the query, retrieves the most similar chunks, and sends them with your question to OpenAI.
4. Use search (Ctrl/Cmd+F) to fuzzy-search across all chats, add favorites, and leave feedback to feed the analytics dashboard.
5. In the demo, explore existing chats and training data to see the workflow end-to-end.

Screenshots

Notes & limits

• The repo does not ship course content; the hosted demo is preloaded with sample web-dev and systems course material. For local runs, upload your own text in the training screen.
• Large uploads are chunked; TOKEN_LIMIT keeps prompts under the OpenAI context window.
• OAuth is optional-if unset, only email/password login shows up.

Next ideas

• Auto-training from mail/forum integrations.
• Sharable conversation links and richer access controls per course.
• One-click cleanup/editing of uploaded training data.