Ozilly | Machine Learning Movie Recommendation Engine

Ozilly is a highly optimized, AI-powered content recommendation system. Built to demonstrate proficiency in Machine Learning, Data Engineering, and Full-Stack deployment, this project leverages vectorization and cosine similarity to deliver highly accurate, personalized cinematic suggestions in real-time.

🟢 Live Deployment: weds2isikahulmykbeldox.streamlit.app

---

🧠 Machine Learning & Architecture

The core recommendation engine relies on Natural Language Processing (NLP) and content-based filtering.

• Feature Engineering & NLP: Processed metadata (genres, keywords, cast, crew) using CountVectorizer / TfidfVectorizer to convert text data into a multidimensional vector space.
• Dimensionality & Similarity Matrix: Computed the Cosine Similarity between movie vectors to mathematically determine content overlap, resulting in a pre-computed 184MB similarity matrix.
• O(1) Matrix Lookups: The Streamlit application loads the pre-computed similarity matrix into memory via pickle, allowing for O(1) index lookups and instant recommendation sorting.

---

🛠 Complete Tech Stack

Data Science & Machine Learning

• Python (3.9+): Core programming language.
• Scikit-Learn: Used for NLP text vectorization and computing the Cosine Similarity matrix.
• Pandas: Utilized for massive dataset cleaning, preprocessing, and DataFrame manipulation.
• NumPy: Employed for high-performance array operations and matrix sorting.

Software Engineering & Backend

• RESTful API Integration: Integrated with the TMDB API to fetch live metadata and high-resolution posters.
• Concurrent Execution: Implemented Python's ThreadPoolExecutor to perform asynchronous, parallel HTTP requests, dropping the API fetching time from ~8 seconds to < 1 second.
• Cloud Data Streaming: Implemented gdown to dynamically stream the massive 184MB machine learning model from Google Drive directly into the app runtime, bypassing GitHub's strict file size limitations (LFS constraints).

Frontend & Deployment

• Streamlit: Engineered the frontend interface with Python, utilizing st.cache_data and st.cache_resource for memoization and efficient state management.
• Custom CSS/HTML Injection: Overrode default Streamlit styling with advanced CSS to create a premium, minimalist (Space Grotesk & Syne typography) user interface.
• Streamlit Community Cloud: Automated CI/CD pipeline for live deployment straight from the GitHub repository.

---

🚀 Running Locally

Prerequisites

• Python 3.9+
• A TMDB API Key

Installation

1. Clone the repository

   git clone https://github.com/yaadhuu/ozilly.git
   cd ozilly

2. Install Dependencies

   pip install -r requirements.txt

3. Run the Application

   streamlit run ozilly.py

   Note: On first initialization, the application will automatically download the required ML similarity model via Google Drive. This ensures your local repository remains lightweight.

---

📝 License

This project is open-source and available under the MIT License.