Building a custom RAG application

As we’ve discussed previously, an RAG application consists of two parts – retrieval and generation. Let’s start with retrieval and take a look at important interfaces exposed by LangChain:

• langchain_core.documents.base.Document: It’s a Serializable object that holds content and metadata. The content is the text of the document (that the downstream chain is dealing with), and the metadata is a collection of key-value pairs that can be associated with the document (you can put filename, size, and any other attributes you’d like there):

  from langchain_core.documents import Document
  doc = Document(page_content="my page",
      metadata={"source_id": "example.pdf", "page": 1})
  print(doc.page_content)

• langchain_core.retrievers.BaseRetriever: It’s an abstract class for retrievers. When you invoke a retriever, it returns a list of Document objects...