About the website

Scrapping from different renowned websites like Aarong, Allen-solly, Apex. Have a Collection of more than 20 Thousand Products 🔥
Hosted the Dataset on HuggingFace Link
A Multimodal Vector database that will support image, text, audio similarity
We have used qdrant for hosting the Vector database on cloud
Every point in the vector database has 4 vectors to support robust and precise multimodal search
We have used OpenAI's text-embedding-3-small model for creating embedding for text. We used CLIP-VIT-B-32 for creating embedding for image. The image embedding supports multimodality. That supports a shared vector space for both text and images. That's how we have both sparse and dense vectors in our vector database
We have used LLAVA-13B model to generate detail specification of product image for the little underdeveloped websites. Then we used the generated description for vectorization purpose
Our services dominantly deals with Smart Search of products. Besides that we also developed review system, recommendation system.
Recommendation system tracks your choice from your cart behaviour and show you recommended related products
Review system is based on sentiment analysis. Your review is categorized so that concurrent users can easily dictate between the positive and negative reviews and also know the reasons
Now Begins our search functionality
Search input is any subset of text, image and audio
We generate corresponding embedding for the input fields.
- text - OpenAI text-embedding-3-small
- Image - CLIP-VIT-B-32
- OpenAI whisper - text - OpenAI text-embedding-3-small
We search into our vector database and fetch the top k relevant hits
We feed the result as payload to gpt for multi level conversation, comparison and so on

Meet the Team

Nazmus Sakib

CSE, BUET

Saad Md Rafid Pial

CSE, BUET

Labid Al Nahiyan Avro

ID:1905110

CSE, BUET