Advanced Autonomous AI Agents (Node.js)

Welcome to the Advanced AI Agents repository. This project explores the transition from standard LLM chat interfaces to fully autonomous, action-oriented AI systems built with JavaScript/Node.js, LangGraph, and LangChain.

The core focus of this repository is to build resilient AI backends capable of complex reasoning, dynamic decision-making, and seamless integration into modern full-stack (MERN) environments.

Architectural Capabilities

• Autonomous Agentic Workflows: Implementing ReAct (Reason + Act) loops where agents independently decide the sequence of actions required to solve complex user queries.
• Dynamic Tool Orchestration: Equipping LLMs with custom, real-world tools, external API execution, and structured output generation validated strictly via Zod.
• Enterprise-Grade RAG Pipelines: Advanced Retrieval-Augmented Generation using semantic search and local embedding models for contextual data retrieval.
• Full-Stack Integration Ready: Modular Node.js architecture designed to be easily plugged into Express.js endpoints for scalable web applications.

Advanced AI Agents (LangChain & RAG) 🚀

This repository contains advanced, production-ready AI Agent architectures. Transitioning from raw Node.js tool-calling loops, this project utilizes LangChain to build scalable AI pipelines, Retrieval-Augmented Generation (RAG) systems, and Multi-Agent workflows.

🛠️ Setup & Installation

This project uses modern JavaScript (ES Modules). To set up the environment on your local machine:

1. Initialize the project:

npm init -y
npm pkg set type="module"

2. Install all dependencies (Includes LangChain, Groq, PDF loaders, and local Hugging Face Embeddings):

npm install @langchain/core @langchain/groq dotenv pdf-parse @langchain/community @langchain/textsplitters @huggingface/transformers langchain@0.1.37 --legacy-peer-deps

3. Environment Variables: Create a .env file in the root directory and add your API keys:

GROQ_API_KEY=your_api_key_here

(Note: Never commit your .env file. Make sure .env and node_modules/ are added to a .gitignore file.)

4. Add your Data: Place a sample PDF file (e.g., example.pdf) in the root directory for the RAG pipeline to process.

---

📂 Directory Structure

• 01_langchain_setup.js

  • Objective: Framework initialization.
  • Details: Sets up the ChatGroq model globally and replaces manual API calls with LangChain's cleaner .invoke() method. Demonstrates standard error handling and model configuration.

• 02_rag_data_prep.js

  • Objective: Data Preparation for RAG.
  • Details: Uses PDFLoader to extract raw text from PDF files. Implements RecursiveCharacterTextSplitter to divide the text into smaller, overlapping chunks (1000 chars) to maintain context and respect LLM token limits.

• 03_rag_embeddings.js

  • Objective: Vector Embeddings.
  • Details: Converts text into numerical vectors using Hugging Face's all-MiniLM-L6-v2 model to understand semantic meaning.

• 04_rag_vector_store.js

  • Objective: Vector Database & Semantic Retrieval.
  • Details: Combines chunking and embeddings to build an in-memory Vector Store (MemoryVectorStore). Performs similaritySearch to mathematically retrieve relevant document sections based on user queries, overcoming exact keyword match limitations.

• 05_rag_chain.js - Retrieval-Augmented Generation (RAG) Architecture Implements a complete RAG pipeline to chat with external documents (PDFs).

  • Core Concepts: Document loading, text splitting, and semantic search.
  • Tech Stack: Uses @langchain/community for local HuggingFace embeddings (all-MiniLM-L6-v2) and in-memory vector stores to provide context-aware responses without relying on outdated LLM training data.

• 06_ai_agent_tools.js - Autonomous ReAct Agent with Tool Orchestration Transitions from static chains to a dynamic, autonomous agent capable of reasoning and taking action.

  • Core Concepts: ReAct (Reason + Act) loop, Tool calling, and strict schema validation.
  • Tech Stack: Powered by @langchain/langgraph and zod. The agent is equipped with custom tools (e.g., Live external API fetchers) and autonomously decides when and how to execute them based on user prompts.

• 07_rag_agent.js - Autonomous RAG Agent (Document Search as a Tool) Integrates a complete Retrieval-Augmented Generation (RAG) pipeline as a dynamically callable tool within the ReAct agent loop.

  • Core Concepts: Agentic search refinement (self-correction/multi-querying), converting dense retrievers into executable tools, and context-aware answer generation.
  • Tech Stack: Combines @langchain/langgraph agent architecture with local HuggingFace embeddings (all-MiniLM-L6-v2) and in-memory vector stores. Utilizes zod to enforce strict query generation schemas for the LLM.

• 08_memory_agent.js - Stateful Agent with Persistent Memory Implements persistent conversational memory across multiple turns, allowing the agent to remember past interactions, user context, and previous tool outputs.

  • Core Concepts: Thread state management, state persistence using graph checkpointers, and conversational history handling without overflowing token limits.
  • Tech Stack: @langchain/langgraph (MemorySaver / Checkpointers), @langchain/core (Message History), and @langchain/groq for context-aware responses.

• 09_structured_streaming.js - Streaming & Structured Output Enforcement Core Concepts: Real-time token streaming (typing effect) and strict JSON schema enforcement for predictable, API-ready LLM outputs. Tech Stack: Uses @langchain/groq and zod for schema definition. Implements .stream() for real-time response delivery and .withStructuredOutput() to guarantee the AI generates strictly formatted JSON data instead of raw Markdown.

• 10_sql_agent.js - Autonomous Text-to-SQL Database Agent Core Concepts: Translating natural language queries into executable SQL commands, querying relational databases dynamically, and synthesizing database outputs into human-readable answers. Tech Stack: Powered by @langchain/groq, typeorm, and mysql. Bridges the gap between traditional MERN stack relational databases and AI by securely generating and executing MySQL queries based on user intent.

• *11_mongo_agent.js - Text-to-MongoDB NoSQL Agent Core Concepts: Converting natural language into complex MongoDB JSON query objects (e.g., .find() parameters), handling NoSQL database connections, and interpreting raw JSON responses into friendly natural language. Tech Stack: Uses @langchain/groq and the native mongodb driver. Integrates seamlessly with the MERN stack's primary database layer to extract insights and analyze collections autonomously.

• 12_router_agent.js - Intelligent Routing & Intent Classification Agent Core Concepts: Analyzing user intent at a high level and dynamically routing queries to the most appropriate sub-system, API, or tool. Acts as the "Traffic Controller" for multi-agent architectures. Tech Stack: Integrates @langchain/groq and @langchain/community/tools/duckduckgo_search. Uses strict JSON-based decision making to classify requests and trigger specific isolated workflows (e.g., Live Weather API, Web Search, or General Chat).

---

🚀 How to Run

To test the semantic search pipeline, ensure your PDF is in the root directory and run:

node filename.js