📘 Lecture notes

About the data

In this lab, we are using a serverless function to import the data required by the agent's tools, into MongoDB. If you want to do this on your own, these datasets are available on Hugging Face:

• mongodb-docs: Markdown versions of a small subset of MongoDB's technical documentation. This dataset is imported into a collection called full_docs.

• mongodb-docs-embedded: Chunked and embedded versions of the articles in the mongodb-docs dataset. This dataset is imported into a collection called chunked_docs.

To learn more about chunking and embedding, here are some resources from our Developer Center:

• How to Choose the Right Chunking Strategy for Your LLM Application
• How to Choose the Best Embedding Model for Your LLM Application

Tool calling

Tool calling, interchangeably called function calling allows an LLM to use external tools such as APIs, databases, specialized machine learning models etc.

In AI agents, an LLM can have access to multiple tools. Given a user query, the LLM decides which tool to invoke and the arguments for the tool call. These arguments are used to execute the tool call and the output is returned back to the LLM to inform its next steps.

The easiest way to define tools in LangChain is using the @tool decorator. The decorator makes tools out of functions by using the function name as the tool name by default, and the function's docstring as the tool's description. The tool call inturn consists of a tool name, arguments, and an optional identifier.

An example of a tool in LangChain is as follows:

@tool("search-tool", return_direct=True)
def search(query: str) -> str:
    """Look up things online."""
    return "MongoDB"

An example of a tool call is as follows:

{
  "name": "search-tool",
  "args": {
    "query": "What is MongoDB?"
  },
  "id": "call_H5TttXb423JfoulF1qVfPN3m"
}

Vector Search in MongoDB

You can learn more about vector search in MongoDB here.