Skip to content

Latest commit

 

History

History
24 lines (17 loc) · 1.12 KB

README.md

File metadata and controls

24 lines (17 loc) · 1.12 KB

Irasutonya

A website to search for Irasutoya pictures with natural languages. Built with Next.js, TailwindCSS and Qdrant.

Features

  • Search pictures with natural language. Model is finetuned on CLIP to get the vectors of pictures. The incoming query string is computed into a vector and Qdrant outputs the pictures with top similarities.

  • (Pseudo) Conversational search. We leverage GPT to rephrase the user's query message to get the proper keyword. Here's an example:

    # Conversation
    [User] I want some pictures of cat.
    [Bot] ...
    [User] Maybe two in the picture?
    
    # Before rephrasing
    [Query] Maybe two in the picture?
    
    # After rephrasing
    [User] two cats in a picture
    

    This feature could be turned on in the Settings - Rephrase (default: false)

  • Diverse reply. We use GPT to generate a "nekoly" reply based on the searched pictures and the previous conversation. This feature could be turned on in the Settings - Diversity (default: false)