Skip to main content

Create embedding for your data via the vector DB

There are two ways to deal with the hallucination from LLM's answer. One is to fine-tune a new model with your domain knowledge. The other one is to build a RAG-based (Retrieval-Augmented Generation) LLM agent via a vector DB. Both solutions can solve the challenge. However, fine-tuning a model requires GPU, which is more expensive. In this series of tutorials, I will walk you through how to build a RAG-based LLM agent.

This article is the first part: create embeddings for your data via the vector DB. This is essential for the chatbot's ability to retrieve relevant information.

Prerequisites​

  1. A GitHub account to log into the flows.network platform. It's free.
  2. An Open API key to convert text to embeddings

Prepare Embeddings for Your Knowledge Base​

There are various methods to split your knowledge base into several chunks. You can choose one to implement. The desired output for this step is a .txt file.

Prepare the source code​

For this part, we already created a repo named demo-RAG-embeddings for you to fork.

Creating the Webhook for Connecting the Vector DB​

Import the forked repository to flows.network, and then the platform will build and deploy the source code. During this process, you will need to add your OpenAI key.

For a detailed guide, you can refer to this video or the Get Started guide.

Upon successful completion of the function build, you will receive a URL at the "Webhook Endpoint" section on the flow details page. This URL will be used to upload the text chunks.

Upload your prepared text chunks to the vector DB​

To add your content to the vector DB, follow these steps:

  • Upload your .txt file to the demo repository you forked earlier. You can save it locally.

  • Open your terminal and execute the following commands:

  # Clone the source code
  git clone https://github.com/your_github_id/demo-RAG-embeddings.git
  cd demo-RAG-embeddings

  # Upload the content
  # Replace the Webhook URL with the one obtained in the previous step
  # The collection_name is up to you, but remember the name for later use
  # "@your_file.txt" should be changed to your file name
  curl "https://code.flows.network/webhook/I9GNgD5hyhFLY25DsOUI?collection_name=your_collection_name&vector_size=1536&reset=1" -X POST --data-binary "@your_file.txt"

After successful addition, the terminal will prompt how many embeddings have been created in total.

Successfully inserted 7 records. The collection now has 7 records in total

Now we have created the embeddings. Next, let's build an LLM agent for your knowledge base.