Academic Resource Platform with Gemini 2.0 AI Quiz Engine

TL;DR: An end-to-end academic platform enabling educators to manage university course resources and generate interactive quizzes automatically from pedagogical materials via Google Gemini 2.0.

1. System Overview & The Vision

Teachers often spend hours manually converting lecture slides and reading materials into assessment questions (multiple-choice, true/false, open-ended). This project addressed the assessment bottleneck by integrating Gemini 2.0 directly into the curriculum workflow.

┌─────────────────────────────────────────────────────────────────────────────┐
│                       DOCUMENT INGESTION & AI PIPELINE                      │
└─────────────────────────────────────────────────────────────────────────────┘

  [Teacher Dashboard]
          │
          │ 1. Upload Course Material (PDF / Markdown / Text)
          ▼
  ┌─────────────────────────────────────────────────────────┐
  │                 Node.js / Express API                   │
  │   - Text Extraction & Token Window Chunking             │
  │   - Prompt Injection Guardrails                         │
  │   - Difficulty & Bloom Taxonomy Framing                 │
  └────────────────────────┬────────────────────────────────┘
                           │
                           │ 2. Structured JSON Output Call
                           ▼
  ┌─────────────────────────────────────────────────────────┐
  │                    Google Gemini 2.0                    │
  │   - Semantic Extraction of Key Concepts                 │
  │   - Distractor Synthesis (Plausible false options)      │
  │   - Explanatory Rationale Generation                    │
  └────────────────────────┬────────────────────────────────┘
                           │
                           │ 3. Validated JSON Schema Response
                           ▼
  ┌─────────────────────────────────────────────────────────┐
  │                     Prisma ORM                          │
  │   - Transactional persistence into PostgreSQL           │
  │   - Draft questions created for teacher review          │
  └─────────────────────────────────────────────────────────┘

2. Technical Architecture

• Frontend Application: Next.js App with Redux Toolkit for complex multi-step quiz state, timer countdowns, and instant visual feedback with Tailwind CSS.
• Backend Service: Express.js REST API structured around clean controller-service-repository patterns.
• Security & Authorization: Role-Based Access Control (RBAC) distinguishing between Teacher (curriculum creators) and Student (quiz takers), secured with cryptographic JWT access tokens.
• Data Modeling (Prisma & PostgreSQL): Relational schema tracking courses, modules, questions, student attempts, score distributions, and AI generation parameters.

3. Gemini 2.0 Integration Mechanics

To ensure questions generated were pedagogically sound and conformed to rigorous JSON structures without hallucinations:

// Strict schema prompting with Gemini 2.0
const prompt = `
You are an expert academic evaluator. Based on the provided lecture transcript, generate 5 questions.
Ensure distractors (incorrect options) reflect common student misconceptions.
Return ONLY valid JSON adhering to the specified schema:
{
  "questions": [
    {
      "stem": "string",
      "type": "MULTIPLE_CHOICE" | "TRUE_FALSE",
      "options": ["string", "string", "string", "string"],
      "correctIndex": number,
      "explanation": "string"
    }
  ]
}
`;

4. Key Takeaways & Results

• Reduced quiz creation time for educators from ~45 minutes to under 20 seconds.
• Delivered an interactive student quiz-taking interface with instant score breakdowns and contextual explanations for missed questions.
• Built a modular, clean codebase demonstrating full-stack engineering proficiency across client state, API boundaries, and LLM orchestration.