07_rag_agent.js

import { ChatGroq } from "@langchain/groq";
import { tool } from "@langchain/core/tools";
import { createReactAgent } from "@langchain/langgraph/prebuilt";
import { z } from "zod";
import "dotenv/config";

//rag related imports
import { RecursiveCharacterTextSplitter } from "@langchain/textsplitters";
import { PDFLoader } from "@langchain/community/document_loaders/fs/pdf";
import { MemoryVectorStore } from "langchain/vectorstores/memory";
import { HuggingFaceTransformersEmbeddings } from "@langchain/community/embeddings/huggingface_transformers";

async function main() {

    //1. Setup RAG
    console.log("\nLoading PDF and setup chunking system..\n");
    const loader = new PDFLoader("./BCA.pdf");
    const docs = await loader.load();
    const textSplitter = new RecursiveCharacterTextSplitter({
        chunkSize: 1000,
        chunkOverlap: 200,
    });
    const chunks = await textSplitter.splitDocuments(docs);

    const embeddings = new HuggingFaceTransformersEmbeddings({
        model: "Xenova/all-MiniLM-L6-v2"
    });

    const vectorStore = await MemoryVectorStore.fromDocuments(chunks, embeddings);
    const retriever = vectorStore.asRetriever({ k: 5 });

    console.log("RAG pipeline ready...\n");

    //2. creating RAG tool
    const ragTool = tool(
        async ({ query }) => {
            try {
                console.log(`[Agent is running Tool] Searching PDF for: "${query}"...\n`);
                //searching the pdf using retriever
                const result = await retriever.invoke(query);
                //Extract text from the retrieved chunks and combine them
                const context = result.map(doc => doc.pageContent).join("\n\n");
                return `found information in pdf : ${context}`
            } catch (error) {
                return `Error : ${error}`;
            }

        },
        {
            name: "search_pdf_document",
            description: "Use this tool to find factual information, syllabus, subjects, or any details from the provided BCA PDF document. Always use this tool when the user asks about the course/college.",
            schema: z.object({
                query: z.string().describe("The exact search query to look for in the document")
            })
        }
    );

    //3. setup lmm (brain)
    const llm = new ChatGroq({
        apiKey: process.env.GROQ_API_KEY,
        model: "qwen/qwen3.6-27b",
        temperature: 0
    });

    //4. combine the llm and the ouur RAG tool
    const agent = createReactAgent({
        llm: llm,
        tools: [ragTool],
    });

    //user question
    const question = "BCA ke first semester me kon kon se subjects hain? List unko bullet points me do."
    console.log(`[User quesion] : ${question}\n`);

    //5. Execution...
    const response = await agent.invoke({
        messages: [
            {
                role: "user",
                content: question
            }
        ]
    });

    console.log(`[Agent response] : ${response.messages[response.messages.length - 1].content}`);

}

main().catch(console.error);