🌌 Ethereal Anime Recommender

A high-performance, full-stack anime discovery platform featuring a hybrid vector search engine (Semantic + Categorical) and a modern React 19 frontend with glassmorphic design.

✨ Core Features

• 🧠 Hybrid Vector Engine: Combines deep semantic understanding (Sentence Transformers) with categorical precision (Genres/Studios) for results that "feel" right.
• ⚡ Real-time Autocomplete: Ultra-fast search suggestions with live metadata as you type.
• 🎭 Genre Intelligence: Dynamic sidebar featuring top-frequency genre filters and immersive content exploration.
• 📺 Netflix-Style UI: Premium hover experiences with pop-out detail windows and high-resolution poster optimization.
• 🚀 Scalable Architecture: Decoupled Flask REST API and Vite-powered React 19 frontend.

---

🛠️ Technology Stack

Frontend

• Framework: React 19 + Vite
• Styling: Tailwind CSS v4
• Icons: Lucide React
• HTTP Client: Axios

Backend

• Framework: Flask (Python 3.11+)
• Vector Search: FAISS (Facebook AI Similarity Search)
• Model: all-MiniLM-L6-v2 (Sentence-Transformers)
• Data Engine: Pandas / NumPy

---

🚀 Getting Started

1. Backend Setup

# Create and activate virtual environment
python -m venv venv
source venv/bin/activate  # venv\Scripts\activate on Windows

# Install dependencies
pip install -r requirements.txt

# Clean data and generate vector indices
# This will create the 'model_artifacts' directory
python data_clean.py

# Launch the API server
python -m backend.run

2. Frontend Setup

cd client

# Install dependencies
npm install

# Start development server
npm run dev

---

📂 Project Structure

├── backend/            # Flask API implementation
├── client/             # React 19 Frontend
├── model_artifacts/    # (Generated) Vector index & processed data
├── data_clean.py       # ML Pipeline: Deduplication & Vectorization
├── scrape_anime.py     # Production-grade scraper
└── anime_data.csv      # Raw dataset

---

📊 Recommendation Methodology

The system uses a Hybrid Indexing strategy:

1. Semantic Layer: Encodes anime synopses into a 384-dimensional vector space using SentenceTransformer.
2. Categorical Layer: One-hot encodes Genres, Studios, and Producers.
3. Similarity: Normalizes and concatenates these layers, then uses FAISS Inner Product (Cosine Similarity) matching to find the closest matches in sub-millisecond time.

📜 License

Licensed under the MIT License.