Food QnA Chatbot : Help Answer Food Related Questions from Your Own Cookbook

Introduction

Retrieval Augmented Generation (RAG) is a powerful technique that allows you to enhance your prompts by retrieving data from external sources and incorporating it into the context. The external data used for augmentation can be gathered from diverse sources such as document repositories, databases, or web search results.

To begin with RAG, you need to convert your documents and user queries into a compatible format to perform relevancy search. This involves converting both the document collection, or knowledge base, and user-submitted queries into numerical representations using embedding. Embedding is a process that assigns numerical values to text, placing them in a vector space.

RAG model architectures then compare the embeddings of user queries with those “vector index” of the knowledge base. By doing so, they identify similar documents in the knowledge base that are relevant to the user’s prompt. These relevant contents from similar documents are appended to the original user prompt.

Finally, the augmented prompt, which now includes the relevant retrieved content, is passed on to the foundation model to generate the final responses. This integration of retrieval and generation significantly improves the quality and relevance of the model’s outputs. Below is an illustrative diagram that demonstrates the overall RAG (Retrieval Augmented Generation) process. .

An Example Project for RAG

I have created an example project to provide a practical demonstration of how RAG works. For more detailed information and insights into the project, you can find comprehensive documentation and additional resources on . This exmaple project will walk you through the RAG process, the data preparation steps, the relevancy search implementation, and how the augmented prompts lead to more contextually, more relevant, and more conversational responses from the foundation model. Please feel free to clone this project on GitHub, and follow the steps below to explore or develope it further.

Set Up Local Virtual Environment

• Step 1: Install Miniconda on MacOS, see instruction here.
• Step 2: Create a default conda env with Python 3.9:

  conda create --name food_qna_app python=3.9 -y
  

• Step 3: Activate the conda env created above: conda activate food_qna_app
• Step 4: Install first set of required libraries in the conda env:

  pip3 install -r requirements.txt
  

• Step 5: Deactivate the conda env when you are done (note: you need activate this virtual env to work on the app)

  conda deactivate 
  

  Set Up Open AI Secrete for Local Dev

  OPENAI_API_KEY will be needed when calling the OpenAI API endpoint for generating embeddings for the documents, so do recommend exporting OPENAI_API_KEY as an enviroment variable on your local dev machine; also, we will need create a secrets.toml file and add the OPENAI_API_KEY there, so the streamlit app can pick up the API key when send requests to OpenAI endpoint during runtime.

• Step 1: Export OPENAI_API_KEY as an enviroment variable on your local dev machine
• Step 2: Create .streamlit directory under the root repo

  cd food_qna_chatbot_demo 
  mkdir .streamlit
  

• Step 3: Create secrets.toml file under .streamlit directory and add your OPENAI_API_KEY there

  OPENAI_API_KEY=<YOUR_OPENAI_API_KEY_HERE>
  

• Step 4: To reduce the risk of accidentally committing your secrets to your repo, add .streamlit to your .gitignore file.

  echo ".streamlit/" >> .gitignore
  

  Set up the Redis Database on your local dev machine

• Step 1: Install Docker Desktop on Mac, and start the docker desktop.
• Step 2: We’re going to use Redis as our database for both document contents and the vector embeddings. You will need the full Redis Stack to enable use of Redisearch, which is the module that allows semantic search - more detail is in the docs for Redis Stack. Run the following command in your terminal to start the docker container:

  docker run -d --name redis-stack -p 127.0.0.1:6379:6379 -p 8001:8001 redis/redis-stack:latest
  

• Step 3: Initiate a Redis connection and create a Hierarchical Navigable Small World (HNSW) index for semantic search using a recipe book which can be found under the directory cook_book_data

  python build_knowledge_base.py
  

  If the recipe book is indexed succefully into the DB, you should expect the following info printed out in the console:

  ===
  Number of documents indexed in the DB: 144
  

  Run the APP

  In your terminal, run the App by

  streamlit run food_qna_app.py
  

  You can start asking questions related to food preparation and cooking, and also some follow up questions. See screenshot below.

Original Content	Chat History - first turn	Chat History - follow up