[Documentation] Basic Tutorials: Vector stores and retrievers

[SkSirius]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

To add a comprehensive documentation page on “Vector Stores and Retrievers” similar to the page found in the Python documentation. This would serve as a basic tutorial aimed at helping users understand and implement vector stores and retrievers within this environment.

Motivation

Creating a consistent tutorial on vector stores and retrievers across various languages/platforms

Proposal (If applicable)

Pull Request: #7133

[SkSirius]

@jacoblee93 Hello! Was digging into vector stores and retrieval and figured the basic tutorial is missing in langchain js docs.

[jacoblee93]

Hey! Yes would be nice to add - although I think we've been thinking about reframing it to address a more concrete problem. Do you have any ideas?

[SkSirius]

How about something like this?
There is a Fake Store API that returns a bunch of products, reviews, and categories.
We can use it to demonstrate a simple RAG that helps search products based on reviews.

// Function to fetch product data from Fake Store API, format it as LangChain Document objects,
// and include metadata for each product (e.g., product ID, category, rating).
const fetchProducts = async () => {
  const response = await fetch("https://fakestoreapi.com/products");
  const data = await response.json();
  return data.map((item: any) => {
    return new Document({
      pageContent: item.description,
      metadata: {
        title: item.title,
        productId: item.id,
        category: item.category,
        rating: item.rating.rate,
      },
    });
  });
};

const chromaConfig = {
  collectionName: "product_reviews",
};

const run = async () => {
  const chatPrompt = ChatPromptTemplate.fromTemplate(`
    Answer this question using the provided context only.
    Question: {question}
    Context: {context}
  `);

  // Fetch product data from the API, format it as documents, and store it in the vector store.
  const documents = await fetchProducts();
  const chroma = await Chroma.fromDocuments(
    documents,
    new OpenAIEmbeddings(),
    chromaConfig
  );
  const retriever = chroma.asRetriever();

  const llm = new ChatOpenAI({ model: "gpt-3.5-turbo" });

  // Define the RAG chain to process the user query through a sequence of actions:
  // 1. Use the retriever to fetch relevant documents (context).
  // 2. Construct a prompt with the retrieved context.
  // 3. Pass the prompt to the LLM to generate a response.
  // 4. Parse the output into a usable string format.
  const ragChain = RunnableSequence.from([
    {
      context: retriever,
      question: new RunnablePassthrough(),
    },
    chatPrompt,
    llm,
    new StringOutputParser(),
  ]);

  const result = await ragChain.invoke(
    "I'm looking for a drive with remarkable speed and transfer capabilities."
  );
  console.log(result);
};

run();