We present a robust catalog searching engine tailored for the dynamic landscape of e-commerce, with the flexibility to extend its architecture for diverse applications. The project revolves around efficiently processing and retrieving information from client-provided CSV files containing detailed product or item information.

• CSV Data Ingestion & Embedding: Stores item data in the Weavite Vector Database for efficient and fast retrieval.

• Unstructured Text Search: Utilizes an inverted index for keyword matching, enabling versatile and precise search capabilities.

• Image Search: Enables image-based product discovery, enhancing the visual exploration of the catalog.

• AI-Powered Filtering: Refines search results based on user intent through intelligent AI-powered filtering.

• Re-Ranking: Prioritizes the most relevant items using advanced re-ranking algorithms.

• Benefits: Provides fast and accurate search capabilities for large catalogs.

  Adaptable to various domains, ensuring versatility in application.

Our project harnesses the capabilities of multiple AI models to elevate search intelligence:

• Embeddings & Re-ranking (all-MiniLM-L6-v2):

  • Captures product meaning and refines search results for optimal relevance.

• Image Summarization (LLaVA):

  • Extracts key visuals from images, enhancing the efficiency of image-based search.

• Planned: Token-based Transformers (Mixtral-8x7B-Instruct-v0.1):

  • Upcoming implementation aimed at further enhancing embeddings and re-ranking.

This AI-powered approach ensures that your searches deliver the results you crave.

• Submit Queries: Enter keywords, phrases, or voice commands to initiate searches.

• Filters and Refinement: Refine search results by applying filters such as category, price, etc. AI-powered suggestions may assist in the refinement process.

• Understanding Results: Results are ranked by relevance and presented in paginated form for user convenience.

• Re-ranking functionality (under development) aims to further enhance result accuracy.

• Upload or Provide Image Link: Upload a product image or provide a link to search for similar items. Image search results are intelligently combined with text-based results.

Note: Re-ranking, Image searching and AI filtering functionalites is currently under development.

  POST /search

  POST /weaviate/schema

   "properties": [
           {
             "name": "unique_id",
             "dataType": ["int"],
             "description": "user id",
             "moduleConfig": {
                 "text2vec-huggingface": {
                 "skip": True,
                 "vectorizePropertyName": True
                 }
             }
             },
             {
             "name": "product_name",
             "dataType": ["text"],
             "description": "product_name",
             "moduleConfig": {
                 "text2vec-huggingface": {
                 "skip": True,
                 "vectorizePropertyName": True
                 }
             }
           }
        ]

  POST /add/data

# This should be the title of each column and type of data
  brand: string,
  category: string,
  market_price: number,
  product_desc: string,
  product_name: string,
  rating: number,
  sale_price: number,
  sub_category: string,
  unique_id: string

Clone the repository

Install Frontend with npm

  cd frontend
  npm install 
  npm run dev

The frontend will be runing on http:https://localhost:3000

Install Backend with Python

  cd backend2
  pip install virtualenv
  virtualenv venv
  venv\Scripts\activate
  pip install --no-cache-dir --requirement ./requirements.txt 
  python main.py

The Server is running on http:https://localhost:5000

MIT