Darajat: AI-Powered Educational Platform

Salam Hackathon Project — AI-powered educational platform for Grades 10-12. Combines RAG-based syllabus search, AI tutoring, and adaptive content generation using LangChain + Google Gemini.

---

🎯 Overview

Darajat (درجات = "Grades" in Arabic) is a hackathon prototype built for the Salam Hackathon. An AI-powered educational platform targeting Grades 10-12 students in Sudan, providing intelligent tutoring, syllabus-aligned content generation, and adaptive learning paths using RAG over curriculum data.

---

🏗️ Architecture

darajat/
├── backend/                    # Bun + Express + TypeScript
│   ├── src/
│   │   ├── routes/
│   │   │   ├── coach.ts        # AI Tutoring endpoint
│   │   │   ├── content.ts      # Content generation
│   │   │   ├── plan.ts         # Learning plan generation
│   │   │   └── index.ts        # Route aggregation
│   │   ├── middleware/
│   │   │   └── loggerMiddleware.ts
│   │   ├── syllabus/           # Curriculum data (JSON)
│   │   │   ├── G10.json
│   │   │   ├── G11.json
│   │   │   └── G12.json
│   │   ├── lib/
│   │   │   ├── llm.ts          # Google Gemini 1.5 Pro config
│   │   │   ├── coachChain.ts   # LangChain RAG chain
│   │   │   ├── contentGenerator.ts
│   │   │   ├── material.ts
│   │   │   ├── planner.ts
│   │   │   └── syllabus.ts
│   │   ├── types/
│   │   │   ├── contentResponse.ts
│   │   │   └── plannerResponse.ts
│   │   └── index.ts            # Express app entry
│   ├── package.json
│   ├── tsconfig.json
│   └── README.md
├── frontend/                    # Static JSON lessons
│   └── Lessons/
│       ├── MathG10.json
│       ├── MathG11.json
│       ├── MathG12.json
│       ├── ScienceG10.json
│       ├── ScienceG11.json
│       └── PhysicsG11.json
├── erd.md                       # Entity Relationship Diagram
├── schema-definition.md        # Database schema
├── package.json                # Root scripts
└── README.md

---

🛠 Tech Stack

Category	Technologies
Runtime	Bun (primary) / Node.js
Language	TypeScript 5+
Framework	Express.js 4.21+
AI/ML	LangChain, LangChain-Google-GenAI, LangChain-Core
LLM	Google Gemini 1.5 Pro (gemini-1.5-pro)
RAG	LangChain Runnables, PromptTemplate
Validation	Zod (planned)
Logging	Custom middleware

---

🚀 Quick Start

git clone https://github.com/oovaa/darajat.git
cd darajat

# Backend
cd backend
bun install
bun run dev    # Starts on http://localhost:3000

# Frontend lessons are static JSON
# Serve with any static server or use frontend framework

---

🔧 Environment Variables

# backend/.env
GOOGLE_API_KEY=your_gemini_api_key
PORT=3000

---

📡 API Endpoints

🎓 Coach (AI Tutor)

POST /coach
Content-Type: application/json

{
  "information": "The first law of thermodynamics states that energy cannot be created or destroyed...",
  "question": "How does the first law apply to closed systems?"
}

Response:

{
  "answer": "The first law of thermodynamics states that in a closed system, the total energy remains constant..."
}

📚 Content Generation (Planned)

POST /content
Content-Type: application/json

{
  "grade": "G11",
  "subject": "Mathematics",
  "topic": "Quadratic Equations",
  "difficulty": "medium"
}

📅 Learning Plan (Planned)

POST /plan
Content-Type: application/json

{
  "grade": "G12",
  "subjects": ["Mathematics", "Physics"],
  "duration_weeks": 12,
  "hours_per_week": 10
}

---

🧠 RAG Pipeline (Coach Chain)

// backend/lib/coachChain.ts
const coach = `
## Context:
{information}

## Question:
{question}

## Task:
Answer the provided question based on the given information

## Response:
`

const coach_prompt = PromptTemplate.fromTemplate(coach)
const chain = RunnableSequence.from([coach_prompt, llm])

export async function coachAnswer(information: string, question: string) {
  const response = await chain.invoke({ information, question })
  return { answer: response.content }
}

Flow:

1. Retrieve relevant syllabus context from G10/G11/G12 JSON
2. Augment prompt with retrieved information
3. Generate answer using Gemini 1.5 Pro
4. Return structured response

---

📚 Curriculum Data (Grades 10-12)

Grade	Subjects	Format
G10	Mathematics, Science	JSON
G11	Mathematics, Science, Physics	JSON
G12	Mathematics, Science	JSON

Sample Structure (backend/src/syllabus/G11.json):

{
  "subjects": [
    "Equations and Functions",
    "Variable Expressions",
    "Order of Operations",
    "Patterns and Equations",
    "Equations and Inequalities",
    "Functions as Rules and Tables",
    "Functions as Graphs",
    ...
  ]
}

---

🛠 Tech Stack

Category	Technologies
Runtime	Bun (primary) / Node.js 20+
Language	TypeScript 5+ (ESM)
Framework	Express.js 4.21+
AI/ML	LangChain, LangChain-Google-GenAI, LangChain-Core
LLM	Google Gemini 1.5 Pro (gemini-1.5-pro)
RAG	LangChain Runnables, PromptTemplate, RunnableSequence
Validation	Zod (planned)
Dev Tools	Bun --watch, TypeScript 5+, ESLint

---

🚀 Quick Start

git clone https://github.com/oovaa/darajat.git
cd darajat

# Backend
cd backend
bun install
bun run dev    # http://localhost:3000

# Frontend lessons (static JSON)
# Serve with any static server or integrate with React/Vue

---

📋 Development Scripts

// backend/package.json
{
  "scripts": {
    "dev": "bun --watch index.ts",
    "start": "bun index.ts"
  }
}

---

🎯 Hackathon Context

Built for Salam Hackathon — addressing educational gaps in Sudan with AI-powered, curriculum-aligned tutoring for Grades 10-12.

Problem

• Limited access to quality tutoring
• Curriculum misalignment with available resources
• No adaptive learning for Sudanese curriculum

Solution

• RAG over Sudanese curriculum (Grades 10-12)
• Gemini 1.5 Pro for high-quality Arabic/English responses
• Modular chains for coaching, content generation, planning
• Offline-first JSON syllabus data

---

📋 Roadmap

• Vector database integration (Qdrant/Pinecone) for better RAG
• Frontend React/Next.js application
• User authentication & progress tracking
• Multi-language support (Arabic/English)
• Offline sync capability
• Teacher dashboard for content management
• Assessment & quiz generation
• Voice interface for accessibility

---

📄 License

MIT License - see LICENSE for details.

---

👤 Author

Omar Abdulrahim
GitHub: @oovaa
Salam Hackathon Participant