ConstVidSearch

ConstVidSearch is an advanced Android application that provides constant-time video search using a vector-based indexing method. This project demonstrates how modern search mechanisms can be implemented efficiently for scalable video datasets.

Features

• Constant-Time Search: Implements O(1) complexity for retrieving relevant videos based on vector embeddings.
• Transformer-Based Embeddings: Uses Microsoft multilingual transformers to generate vector embeddings for video titles and descriptions.
• Cloud Storage: Video files and metadata are stored securely on AWS services like S3 and DynamoDB.
• Scalable Architecture: Designed to handle large-scale video datasets using Pinecone Vector Database for fast and accurate retrieval.
• User-Friendly Interface: Allows users to search videos by title and view results instantly.

Architecture Overview

Preprocessing Phase

1. User Input: Users upload a video with a title, description, and date.
2. Storage:
   • The video, thumbnail, and metadata are stored in an AWS S3 Bucket.
   • Metadata (e.g., title, description, thumbnail URL, video URL) is stored in DynamoDB.
3. Embedding Generation:
   • Titles and descriptions are transformed into vector embeddings using Microsoft multilingual transformers.
   • The embeddings are indexed into Pinecone Vector Database for fast retrieval.

Searching Phase

1. User Query: The user searches for a video by providing a title.
2. Vector Matching:
   • The query title is transformed into a vector using the same transformer model.
   • Pinecone Vector Database performs a nearest neighbor search to retrieve top K matching results.
3. Data Retrieval:
   • Metadata of the top results is fetched from DynamoDB using partition keys.
   • Results (title, description, thumbnail, video URL) are displayed on the user interface.

Tech Stack

• Android: Java-based mobile application development.
• AWS Services: S3 (video storage), DynamoDB (metadata storage).
• Transformer Models: Microsoft multilingual transformer.
• Vector Database: Pinecone for fast and scalable vector search.

How to Use

1. Upload a Video:
   • Enter the video title, description, and date.
   • Upload the video file and thumbnail.
2. Search a Video:
   • Use the search bar to enter a video title.
   • View the search results with relevant video thumbnails and descriptions.

Screenshots

Preprocessing Flow

Searching Flow

Installation

1. Clone the repository:

   git clone https://github.com/your-username/ConstVidSearch.git

2. Open the project in Android Studio.
3. Configure your AWS credentials in the app settings.
4. Build and run the application on an Android device/emulator.

Future Enhancements

• Add support for advanced filters like tags, categories, and timestamps.
• Implement personalized recommendations based on user history.
• Support multilingual search queries.

Contributing

Contributions are welcome! Please fork the repository, create a feature branch, and submit a pull request.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Contact

• GeeksforGeeks Profile: Your gfg
• LinkedIn: Your LinkedIn
• GitHub: Your GitHub

---

Feel free to explore, contribute, and give feedback on the project. Together, let's make video search faster and smarter! 🚀