PrepifyAI — AI-Powered Exam Preparation Platform

Final Year Project — an AI study assistant for Pakistani students preparing for FBISE Matric/FSc board exams and entry tests (MDCAT, ECAT).

PrepifyAI turns a library of textbooks and past papers into a personalized prep experience: it generates practice questions grounded in real syllabus content, predicts likely exam topics from years of past papers, answers student questions with a context-aware chatbot, and adapts what each student practices next based on their performance.

---

✨ Key Features

Area	What it does
🧠 AI question generation	Retrieval-augmented generation (RAG) over textbook content produces grounded MCQs, short, and long questions with answer keys.
🔮 Topic prediction	A fine-tuned DistilBERT model predicts likely exam topics by analyzing the full corpus of past papers.
💬 Study chatbot	A context-aware assistant that answers concept questions and produces structured performance feedback.
📊 Adaptive practice	Recommends the next best questions and a revision plan from each student's strengths and weak topics.
📄 Past paper pipeline	OCR + parsing extracts questions from scanned past-paper PDFs and embeds them for semantic search.
🏆 Performance & gamification	Analytics dashboards, progress tracking, streaks, and points.
🛠️ Admin tools	Upload textbooks/past papers, manage the catalog, monitor prediction accuracy and system health.

Deep dive: the AI/ML system is documented separately in docs/AI_ARCHITECTURE.md.

---

🏗️ Architecture

┌──────────────────────────┐        REST/JSON         ┌───────────────────────────┐
│   Mobile + Web Frontend  │  ───────────────────▶    │      FastAPI Backend      │
│  React Native (Expo)     │   JWT auth, async API    │  layered: routes →        │
│  expo-router, TypeScript │  ◀───────────────────    │  services → repositories  │
└──────────────────────────┘                          └─────────────┬─────────────┘
                                                                     │
                          ┌──────────────────────────────────────────┼───────────────────────────┐
                          ▼                          ▼                ▼                             ▼
                  ┌───────────────┐        ┌──────────────────┐  ┌─────────────┐         ┌──────────────────┐
                  │ PostgreSQL +  │        │ FAISS + Sentence │  │ DistilBERT  │         │  Groq LLM        │
                  │ pgvector      │        │ Transformers     │  │ (topic pred)│         │ (Llama 3.1/3.3)  │
                  │ (app data +   │        │ (RAG retrieval)  │  │             │         │ question gen +   │
                  │ embeddings)   │        │                  │  │             │         │ chatbot          │
                  └───────────────┘        └──────────────────┘  └─────────────┘         └──────────────────┘

Backend is layered for testability: routes/ (HTTP) → services/ (business + AI logic) → repositories/ (data access) → models/ (SQLAlchemy), with schemas/ (Pydantic) for validation and core/ for config, security, and caching.

---

🧰 Tech Stack

Frontend

• React Native + Expo (SDK 54), expo-router (file-based routing)
• TypeScript, NativeWind / Tailwind, Recharts
• Runs on Android, iOS, and Web

Backend

• FastAPI (async) + Uvicorn
• PostgreSQL with pgvector for 384-d embeddings
• SQLAlchemy 2.0 (async) + Alembic migrations
• JWT auth (python-jose) + bcrypt, role-based access control

AI / ML

• Groq (Llama 3.1 8B / 3.3 70B) for generation and chat
• sentence-transformers (all-MiniLM-L6-v2) + FAISS for RAG retrieval
• Hugging Face Transformers — fine-tuned DistilBERT for topic prediction
• PyTesseract + pdf2image / PyPDF2 for OCR of past-paper PDFs

Tooling

• Docker Compose (pgvector Postgres), GitHub Actions CI, pytest suite

---

📁 Repository Layout

PrepifyAI/
├── FYP-Backend-main/        # FastAPI backend + AI/ML services
│   ├── app/
│   │   ├── routes/          # API endpoints
│   │   ├── services/        # business logic + AI (RAG, prediction, chatbot, OCR)
│   │   ├── repositories/    # data access
│   │   ├── models/          # SQLAlchemy models
│   │   ├── schemas/         # Pydantic schemas
│   │   ├── core/            # config, security, caching
│   │   └── alembic/         # database migrations
│   ├── scripts/             # ops + manual check scripts
│   └── tests/               # pytest suite (CI-run)
├── FYP_FRONTEND-main/       # React Native / Expo app
│   ├── app/                 # expo-router screens
│   └── src/                 # screens, services, components, context
├── docs/AI_ARCHITECTURE.md  # AI/ML deep dive
└── docker-compose.yml       # database + cache for local dev

---

🚀 Quick Start

Full backend setup, Windows notes, and Android-emulator troubleshooting live in FYP-Backend-main/README.md.

1. Database

docker compose up -d
docker exec prepifyai_postgres psql -U postgres -d PrepifyAI_Main \
  -c "CREATE EXTENSION IF NOT EXISTS vector;"

2. Backend

cd FYP-Backend-main
python -m venv venv && source venv/bin/activate     # Windows: venv\Scripts\activate
pip install -r app/requirements.txt
cp app/.env.example app/.env                          # add DATABASE_URL, SECRET_KEY, GROQ_API_KEY, ...
uvicorn app.main:app --reload --host 0.0.0.0 --port 8001

API docs: http://localhost:8001/docs

3. Frontend

cd FYP_FRONTEND-main
npm install
npm run dev            # Expo; press a (Android), i (iOS), or w (web)

Point the app at the API via EXPO_PUBLIC_API_BASE_URL / EXPO_PUBLIC_API_PORT (and EXPO_PUBLIC_ANDROID_EMULATOR_HOST for emulators).

---

✅ Testing

cd FYP-Backend-main
python -m pytest tests/ -v          # automated suite (also runs in CI)

The suite covers auth & RBAC, past-paper upload/extraction, question management, performance tracking, and prediction-service loading.

---

📌 Notes

• Secrets live only in app/.env (never committed); see app/.env.example.
• The DistilBERT weights (pretrained/) and large datasets are git-ignored — question generation gracefully falls back when the heavy stack is absent.