Try now: https://me-shubham-hf-test.hf.space/

Recipe Finder

Project structure:

• get_data.ipynb: Fetches recipe data from upliance.ai. and saves it into json files:
  • recipe_category.json
  • recipe_ingredients.json
• prepare_data.ipynb: Prepares recipe data into tabular format. Creates document for each recipe. And, saves final table in recipe.parquet
• vector_db.ipynb: Create chromaDB vector store in vector-store folder and saves the recipe document embeddings in it.
• app: FastAPI webserver.

How to run: (Be patient it might take time when you run it for the first time, as it downloads the model.)

• Install python 3.10
• create virtual environment: python -m venv venv
• activate virtual environment:
  • windows: source ./venv/Scripts/activate
  • linux/mac: source ./venv/bin/activate
• pip install -r requirements.txt
• [optional]: run prepare data and vector store bash script.
  • Not required if recipe.parquet and vector-store already present.
  • chmod +x script.sh
  • ./script.sh
• run FastAPI server: uvicorn app.main:app --port 8080
• visit http://127.0.0.1:8080/get-recipe-data

Resources Used:

• app/response.html for frontend was entirely generated via Claude and Mistral AI.
• Libraries used: pandas, numpy, pyarrow, requests, chromadb, sentence-transformers, torch, fastapi, jupyter, ipykernel

Scope of Improvement:

• Spelling mistakes in query might result not getting relevant results.
• This solution might not work well for hindi or Hinglish text, because the model used was trained on English language data.

<iframe src="https://me-shubham-hf-test.hf.space" frameborder="0" width="850" height="450" ></iframe>