Let’s write some code!

We’ll create a simple QA chat bot that indexes a PDF and a youtube video as it’s knowledge base. This will allow accurate answers.

Code Explanation:

1. Imports: The code starts by importing necessary modules for configuration, vector storage, embeddings, document loading, text splitting, and the OpenAI SDK.
2. Setting Up Variables:

const question = process.argv[2] || 'hi'
const video = `https://youtu.be/zR_iuq2evXo?si=cG8rODgRgXOx9_Cn`

• question: Grabs the first command-line argument as the question to be queried. If none is provided, defaults to 'hi'.
• video: Defines a YouTube video URL.

1. Create Memory Vector Store:

export const createStore = (docs) =>
  MemoryVectorStore.fromDocuments(docs, new OpenAIEmbeddings())

• This function initializes a memory vector store using the provided documents and OpenAI embeddings.

1. Load and Split YouTube Video Text:

export const docsFromYTVideo = async (video) => {
  const loader = YoutubeLoader.createFromUrl(video, {
    language: 'en',
    addVideoInfo: true,
  })
  return loader.loadAndSplit(
    new CharacterTextSplitter({
      separator: ' ',
      chunkSize: 2500,
      chunkOverlap: 100,
    })
  )
}

• This function takes a YouTube video URL, loads its content, and splits it into manageable chunks. It uses YoutubeLoader to fetch the video's content and CharacterTextSplitter to divide the content.

1. Load and Split PDF Text:

export const docsFromPDF = () => {
  const loader = new PDFLoader('xbox.pdf')
  return loader.loadAndSplit(
    new CharacterTextSplitter({
      separator: '. ',
      chunkSize: 2500,
      chunkOverlap: 200,
    })
  )
}

• Similar to the YouTube function, this one loads content from a PDF file and splits it into chunks.

1. Load Combined Document Store:

const loadStore = async () => {
  const videoDocs = await docsFromYTVideo(video)
  const pdfDocs = await docsFromPDF()

  return createStore([...videoDocs, ...pdfDocs])
}