langchain-rag (JavaScript/TypeScript)

Overview

Retrieval Augmented Generation (RAG) enhances LLM responses by fetching relevant context from external knowledge sources. Instead of relying solely on training data, RAG systems retrieve documents at query time and use them to ground responses.

Key Concepts:

• Document Loaders: Ingest data from files, web, databases
• Text Splitters: Break documents into chunks
• Embeddings: Convert text to vectors
• Vector Stores: Store and search embeddings
• Retrievers: Fetch relevant documents for queries

RAG Pipeline

1. Index: Load → Split → Embed → Store
2. Retrieve: Query → Embed → Search → Return docs
3. Generate: Docs + Query → LLM → Response

Decision Tables

Vector Store Selection

Store	When to Use	Why
MemoryVectorStore	Development, testing	In-memory, fast, ephemeral
Chroma	Local production	Persistent, open-source
Pinecone	Cloud, scale	Managed, fast, scalable
Faiss	High performance	Fast similarity search

Embedding Model Selection

Model	When to Use	Dimension
text-embedding-3-small	Cost-effective	1536
text-embedding-3-large	Best quality	3072
text-embedding-ada-002	Legacy	1536

Code Examples

Basic RAG Setup

import { ChatOpenAI, OpenAIEmbeddings } from "@langchain/openai";
import { MemoryVectorStore } from "@langchain/classic/vectorstores/memory";
import { RecursiveCharacterTextSplitter } from "@langchain/textsplitters";

// 1. Load documents (example: in-memory text)
const docs = [
  { pageContent: "LangChain is a framework for building LLM applications.", metadata: {} },
  { pageContent: "RAG stands for Retrieval Augmented Generation.", metadata: {} },
];

// 2. Split documents
const splitter = new RecursiveCharacterTextSplitter({
  chunkSize: 500,
  chunkOverlap: 50,
});
const splits = await splitter.splitDocuments(docs);

// 3. Create embeddings and store
const embeddings = new OpenAIEmbeddings({
  model: "text-embedding-3-small",
});

const vectorStore = await MemoryVectorStore.fromDocuments(splits, embeddings);

// 4. Create retriever
const retriever = vectorStore.asRetriever(4); // Top 4 results

// 5. Use in RAG
const model = new ChatOpenAI({ model: "gpt-4.1" });

const query = "What is RAG?";
const relevantDocs = await retriever.invoke(query);

const context = relevantDocs.map(doc => doc.pageContent).join("\n\n");
const response = await model.invoke([
  { role: "system", content: `Use the following context to answer questions:\n\n${context}` },
  { role: "user", content: query },
]);

console.log(response.content);

Loading Web Pages

import { CheerioWebBaseLoader } from "@langchain/community/document_loaders/web/cheerio";

const loader = new CheerioWebBaseLoader(
  "https://docs.langchain.com/oss/javascript/langchain/agents"
);

const docs = await loader.load();
console.log(`Loaded ${docs.length} documents`);

Loading PDF Files

import { PDFLoader } from "@langchain/community/document_loaders/fs/pdf";

const loader = new PDFLoader("./document.pdf");
const docs = await loader.load();

Advanced Text Splitting

import { RecursiveCharacterTextSplitter } from "@langchain/textsplitters";

const splitter = new RecursiveCharacterTextSplitter({
  chunkSize: 1000,        // Characters per chunk
  chunkOverlap: 200,      // Overlap for context continuity
  separators: ["\n\n", "\n", " ", ""],  // Split hierarchy
});

const splits = await splitter.splitDocuments(docs);

Using Chroma (Persistent)

import { Chroma } from "@langchain/community/vectorstores/chroma";
import { OpenAIEmbeddings } from "@langchain/openai";

const embeddings = new OpenAIEmbeddings();

// Create and populate
const vectorStore = await Chroma.fromDocuments(
  splits,
  embeddings,
  { collectionName: "my-docs" }
);

// Later: Load existing
const vectorStore2 = await Chroma.fromExistingCollection(
  embeddings,
  { collectionName: "my-docs" }
);

Advanced Retrieval

// Similarity search with scores
const results = await vectorStore.similaritySearchWithScore(query, 5);
for (const [doc, score] of results) {
  console.log(`Score: ${score}, Content: ${doc.pageContent}`);
}

// MMR (Maximum Marginal Relevance) for diversity
const retriever = vectorStore.asRetriever({
  searchType: "mmr",
  searchKwargs: { fetchK: 20, lambda: 0.5 },
  k: 5,
});

Metadata Filtering

// Add metadata when creating documents
const docs = [
  {
    pageContent: "Python programming guide",
    metadata: { language: "python", topic: "programming" }
  },
  {
    pageContent: "JavaScript tutorial",
    metadata: { language: "javascript", topic: "programming" }
  },
];

// Search with filter
const results = await vectorStore.similaritySearch(
  "programming",
  5,
  { language: "python" }  // Only Python docs
);

RAG with Agent

import { createAgent } from "langchain";
import { tool } from "langchain";
import { z } from "zod";

const searchDocs = tool(
  async ({ query }) => {
    const docs = await retriever.invoke(query);
    return docs.map(d => d.pageContent).join("\n\n");
  },
  {
    name: "search_docs",
    description: "Search documentation for relevant information",
    schema: z.object({
      query: z.string().describe("Search query"),
    }),
  }
);

const agent = createAgent({
  model: "gpt-4.1",
  tools: [searchDocs],
});

const result = await agent.invoke({
  messages: [{ role: "user", content: "How do I create an agent?" }],
});

Hybrid Search (Keywords + Semantic)

// Combine keyword and vector search
import { similarity } from "ml-distance";

async function hybridSearch(query: string, k: number = 5) {
  // Vector search
  const vectorResults = await vectorStore.similaritySearch(query, k);
  
  // Keyword search (simple example)
  const allDocs = await vectorStore.getAllDocuments();
  const keywordResults = allDocs.filter(doc =>
    doc.pageContent.toLowerCase().includes(query.toLowerCase())
  );
  
  // Combine and deduplicate
  const combined = [...vectorResults, ...keywordResults];
  const unique = Array.from(new Set(combined.map(d => d.pageContent)))
    .map(content => combined.find(d => d.pageContent === content));
  
  return unique.slice(0, k);
}

Boundaries

What You CAN Configure

✅ Chunk size/overlap: Control document splitting
✅ Embedding model: Choose quality vs cost
✅ Number of results: Top-k retrieval
✅ Metadata filters: Filter by document properties
✅ Search algorithms: Similarity, MMR, hybrid

What You CANNOT Configure

❌ Embedding dimensions (per model): Fixed by model
❌ Perfect retrieval: Semantic search has limits
❌ Real-time document updates: Re-indexing needed

Gotchas

1. Forgetting to Split Documents

// ❌ Problem: Entire documents are too large
await vectorStore.addDocuments(largeDocs);  // May hit token limits

// ✅ Solution: Always split first
const splits = await splitter.splitDocuments(largeDocs);
await vectorStore.addDocuments(splits);

2. Chunk Size Too Small/Large

// ❌ Problem: Too small - loses context
const splitter = new RecursiveCharacterTextSplitter({ chunkSize: 50 });

// ❌ Problem: Too large - hits limits
const splitter = new RecursiveCharacterTextSplitter({ chunkSize: 10000 });

// ✅ Solution: Balance (500-1500 typically good)
const splitter = new RecursiveCharacterTextSplitter({
  chunkSize: 1000,
  chunkOverlap: 200,
});

3. No Overlap

// ❌ Problem: No overlap - context breaks at boundaries
const splitter = new RecursiveCharacterTextSplitter({
  chunkSize: 1000,
  chunkOverlap: 0,  // Bad!
});

// ✅ Solution: Use overlap (10-20% of chunk size)
const splitter = new RecursiveCharacterTextSplitter({
  chunkSize: 1000,
  chunkOverlap: 200,  // 20%
});

4. Not Persisting Vector Store

// ❌ Problem: Using MemoryVectorStore in production
const vectorStore = await MemoryVectorStore.fromDocuments(docs, embeddings);
// Lost on restart!

// ✅ Solution: Use persistent store
const vectorStore = await Chroma.fromDocuments(
  docs,
  embeddings,
  { collectionName: "prod-docs" }
);

Links to Documentation

• RAG Tutorial
• Document Loaders
• Text Splitters
• Vector Stores
• Embeddings