SchoolMate – AI Tutor & Quiz Platform

SchoolMate is an AI-powered learning platform that enables students to chat with their PDFs, generate quizzes, and engage in a 1:1 AI Tutor experience.

The goal is not only to answer questions but also to develop critical thinking skills by allowing students to debate with the AI, ask follow-up questions, and reflect on their readings.

---

🚀 Features

👩‍🏫 Learning Experience

• Chat with PDF Documents

  • Upload lecture slides, textbooks, or notes (in PDF format).
  • Ask questions in natural language.
  • AI provides answers based on the content of the uploaded document using RAG (Retrieval-Augmented Generation).

• Two AI Modes

  • Tutor Mode
    • Socratic-style AI Tutor.
    • Engages students with probing questions and challenges their assumptions.
    • Aims to improve critical thinking rather than just recall information.
  • Summary Mode
    • Provides concise explanations and key-point summaries.
    • Ideal for quick revisions and review.

📚 Quiz Generation

• Create quizzes based on:

  • PDF documents to reinforce key concepts and knowledge.
  • Web links to test understanding of online materials.
  • Plain text or notes for personalized quizzes.

• Features of the Quiz System

  • Generate quizzes directly from the contents of uploaded PDFs, allowing users to focus on specific sections or topics.
  • Save quizzes to a personal Quiz Library for later review and practice.
  • Support for viewing past quiz attempts to track progress over time.
  • Multiple-choice, true/false, and open-ended questions formats for varied assessment styles.

• Quiz Library

  • Users can view, edit, and delete quizzes.
  • Retry quizzes to assess improvement and mastery of content.

---

🧠 AI & RAG

• Utilizes Retrieval-Augmented Generation (RAG):
  1. Extracts text from uploaded PDFs.
  2. Creates embeddings and stores them in a vector database.
  3. For each user question, retrieves the most relevant content segments.
  4. Sends the retrieved context and user messages to the AI model for generating responses.
• Built on top of AI SDK with Google Gemini models.
• Uses streaming responses for smooth user interactions.

---

🛠 Tech Stack

Frontend

• Next.js (App Router)
• TypeScript
• React
• Tailwind CSS
• shadcn/ui
• React Markdown (for rendering AI responses)
• TanStack Query (React Query) – for client-side caching and server state

Backend & Infrastructure

• Next.js API Routes
• Prisma (ORM)
• PstgreSQL
• Pinecone Vector store for embeddings
• Vercel Blob File storage for PDFs

Authentication

• Clerk – for authentication and user management

AI

• AI SDK (ai, @ai-sdk/react)
• Google Gemini API via @ai-sdk/google
• Custom system prompts for Tutor and Summary modes
• LangChain for text split to chunk before embeddings

---

📁 Main Features in Code (Quick Tour)

Chat API (/api/chat)

• Handles chat requests from the client.

• Expects a request body:

  type ChatRequestBody = {
  	messages: UIMessage[];
  	documentId?: string;
  	chatId?: string;
  	mode?: "tutor" | "summary";
  };

• Flow:

  1. Authenticate the user (Clerk).
  2. Extract the user's latest question from messages.
  3. Query the vector store with embeddings.embedQuery(question) and getVectorStore().
  4. Build context from the top matches.
  5. Select the system prompt based on the mode:
     • Summary Mode → Direct, concise answers.
     • Tutor Mode → Interactive, challenges the user, promotes critical thinking.
  6. Call streamText with Google Gemini model.
  7. On finish, save chat and messages to the database via Prisma.

Chat Page (/chat/[documentId])

• Layout:

  • Left Sidebar – Document info, page navigation.
  • Center – PDF viewer (with scrolling, zoom, and page synchronization).
  • Right Panel – AI chat UI (messages, input, mode switcher, history).

• Uses useChat from @ai-sdk/react with a custom DefaultChatTransport.

Document Management

• /api/documents:
  • List user’s uploaded documents.
  • Delete a document.
• Upload flow:
  • PDF file upload.
  • Store file.
  • Extract text, generate chunks, embed, and store in the vector database.
  • Save document metadata to the database.

Quiz System

• Quiz creation pages:

  • From PDF upload (/quiz/create/pdf-upload)
  • From URL (/quiz/create/insert-link)
  • From text input (/quiz/create/text)

• /quiz – Quiz library for user access:

  • View a list of created quizzes with search and filter options.
  • Access quiz details and attempt history.

---

⚙️ Getting Started

1. Prerequisites

• Node.js (v18+ recommended)
• pnpm / npm / yarn
• PostgreSQL (or any DB supported by Prisma)
• Clerk account (for authentication)
• Google Generative AI API key
• Vector DB (or your own implementation wrapped in getVectorStore)

2. Environment Variables

Create a .env file in the project root:

# Database
DATABASE_URL="postgresql://USER:PASSWORD@HOST:PORT/DB_NAME"

# Clerk
CLERK_PUBLISHABLE_KEY="pk_..."
CLERK_SECRET_KEY="sk_..."

# Google Gemini (via AI SDK)
GOOGLE_GENERATIVE_AI_API_KEY="..."

# Any additional vector DB / storage configs you use
# VECTOR_DB_URL=...
# FILE_STORAGE_BUCKET=...

3. Install Dependencies

pnpm install
# or
npm install
# or
yarn install

4. Database Setup

npx prisma migrate dev
# or
npx prisma db push

(Optional) Open Prisma Studio to inspect data:

npx prisma studio

5. Run the Development Server

pnpm dev
# or
npm run dev
# or
yarn dev

Open your browser and go to http://localhost:3000.

---

🧩 Project Structure (Simplified)

src/
  app/
    chat/
      [documentId]/
        page.tsx            # Chat page (PDF + AI chat + mode switch)
    api/
      chat/
        route.ts            # Chat API (RAG + AI Tutor / Summary mode)
      documents/
        route.ts            # List / upload documents
        [id]/
          route.ts          # Delete document
      quiz/
        route.ts            # Quiz CRUD operations
  components/
    chat/
      ChatLeftSidebar.tsx
      ChatPdfViewer.tsx
      ChatRightPanel.tsx
    quiz/
      QuizCard.tsx
    AuthRequireCard.tsx
    PdfMainViewer.tsx
  lib/
    prisma.ts               # Prisma client
    vector-store.ts         # Embeddings and vector store helpers
    actions/
      chat.actions.ts       # Functions for fetching chat data
  types/
    types.global.ts         # Shared types (ChatMode, ChatSession, DocumentItem, ...)

---

🧪 Key Concepts

• Socratic Tutoring

  • Tutor mode encourages the AI not to provide direct answers.
  • It asks questions to lead the student to discover answers themselves.

• RAG (Retrieval-Augmented Generation)

  • Keeps AI responses grounded in the uploaded document.
  • Significantly reduces inaccuracies by always sourcing context from the vector database.

• Mode-aware Prompting

  • Same backend flow, but different interaction styles:
    • Summary mode delivers fast, factual responses.
    • Tutor mode engages users in interactive discussions.

---

📝 License

This is a personal learning project.
Feel free to use it as a reference or inspiration for your own learning platform.