rag-api

A simple TypeScript/Node.js package for building RAG (Retrieval-Augmented Generation) pipelines with LangChain, Pinecone, and OpenAI models.

It provides two main functions:

• createEmbeddings: preprocesses and stores book documents (PDF) into a Pinecone vector store.
• buildChat: creates a conversational retrieval QA chain with memory, LLM, and retriever to chat with the ingested book.

🚀 Installation

npm install rag-api

or with Yarn:

yarn add rag-api

📦 Exports

1. createEmbeddings(args: CreateEmbeddingsArgs)

Loads a PDF, splits it into chunks, and stores embeddings in Pinecone.

Arguments (CreateEmbeddingsArgs):

interface CreateEmbeddingsArgs {
  bookId: number;      // unique identifier for the book
  bookPath: string;    // local path to the PDF file
}

Example:

import { createEmbeddings } from "rag-api";

await createEmbeddings({
  bookId: 1,
  bookPath: "./books/mybook.pdf",
});

2. buildChat(args: ChatArgs)

Creates a conversational retrieval QA chain for querying the book.

Arguments (ChatArgs):

interface ChatArgs {
  conversationId: number;   // unique conversation/session ID
  llmTemperature: number;   // temperature for LLM responses
  bookId: number;           // ID of the book previously embedded
  streaming: boolean;       // enable/disable streaming responses
  databaseUtils?: DatabaseUtils; // optional DB utils for persisting messages
}

interface DatabaseUtils {
  createMessage: (message: IMessage, conversationId: number) => any;
  getMessagesByConversationId: (conversationId: number) => Promise<(AIMessage | HumanMessage | SystemMessage)[]>;
}

Example:

import { buildChat } from "rag-api";

const chatChain = buildChat({
  conversationId: 42,
  llmTemperature: 0.7,
  bookId: 1,
  streaming: false,
});

// later, you can use the chain to ask questions
const response = await chatChain.call({ question: "Summarize chapter 2" });
console.log(response);

⚡ Requirements

• OpenAI API Key (used by ChatOpenAI)
• Pinecone API Key & Environment (used for storing embeddings)

Make sure you set these environment variables:

export OPENAI_API_KEY="your_openai_api_key"
export PINECONE_API_KEY="your_pinecone_api_key"
export PINECONE_ENVIRONMENT="your_environment"

📖 Workflow

1. Run createEmbeddings with a PDF to index it into Pinecone.
2. Use buildChat with the same bookId to start a conversational QA over the indexed content.
3. Optionally provide databaseUtils to persist conversations in your own database.

🛠 Development

Clone the repo and install dependencies:

git clone https://github.com/yourusername/rag-api.git
cd rag-api
npm install

Build:

npm run build

📄 License

MIT