Skip to content

README.md

Latest commit

 

History

History
137 lines (105 loc) · 4.85 KB

README.md

File metadata and controls

137 lines (105 loc) · 4.85 KB
Raw

AI-Powered Inventory & Knowledge Graph Chatbots

This repository contains a dual-agent system designed to interact with enterprise data using natural language. It demonstrates two distinct approaches to AI data retrieval and management: querying a relational SQLite database (Inventory Bot) using LangGraph, and managing a Neo4j graph database (Knowledge Graph Agent) using LlamaIndex.

🧠 System Overview

The project is split into two primary terminal-based chatbots, each optimized for a specific type of database architecture. Additionally, an experimental Jupyter Notebook is included for Gemini-powered SQL generation via FastAPI.

1. Inventory Chatbot (Relational SQL Agent)

An AI agent built with LangGraph that translates natural language into SQLite queries. It features a robust self-correction routing loop that automatically catches and fixes syntactically incorrect SQL queries before returning the final answer.

✨ Features

  • Natural Language to SQL: Converts questions about assets, locations, and vendors into executable SQLite code.
  • Self-Correction Loop: Uses an AI corrector node to detect execution errors and regenerate the query autonomously.
  • Structured State Management: Uses LangGraph's StateGraph to pass the user question, generated SQL, execution results, and errors between isolated nodes.

🏗️ Architecture Diagram

graph TD
    A[User Input] -->|Question| B(Generator Node)
    B -->|Generates SQL| C(Executor Node)
    C -->|Runs Query on SQLite| D{Execution Error?}
    
    D -- Yes --> E(Corrector Node)
    E -->|Fixes SQL Syntax| C
    
    D -- No --> F(Responder Node)
    F -->|Synthesizes Results| G[Final NL Output to User]

    subgraph LangGraph State Workflow
    B
    C
    E
    F
    end
Loading

2. Knowledge Graph Agent (Neo4j Agent)

A complete CRUD-capable agent that allows users to manage facts, entities, and relationships within a Neo4j graph database using natural language commands.

✨ Features

  • Two-Step Intent & Entity Classification: Precisely identifies if the user wants to add, inquire, edit, or delete data, and then dynamically extracts the required entities (entity, relation, value).
  • Dynamic Cypher Translation: Automatically generates and executes complex Cypher queries (using MERGE, MATCH, and conditional FOREACH loops for orphaned node deletion).
  • Natural Synthesis: Provides human-readable summaries of the database actions via a dedicated LlamaIndex Synthesis Engine.

🏗️ Architecture Diagram

graph TD
    A[User Input] --> B(Intent Classifier)
    B -->|add, inquire, edit, delete| C(Entity Extractor)
    C -->|JSON: entity, relation, value| D(Cypher Engine)
    
    D <-->|Executes Structured Query| E[(Neo4j AuraDB)]
    
    D -->|Raw Database Result| F(Synthesis Engine)
    F -->|Human-Readable Summary| G[Final NL Output to User]
    
    subgraph LlamaIndex & Groq Workflow
    B
    C
    D
    F
    end
Loading

🚀 Installation & Setup

Prerequisites

  • Python 3.10+
  • SQLite3
  • Neo4j Database (Local or AuraDB)
  • API Keys (Groq / OpenAI)

1. Clone the Repository

git clone [https://github.com/MalakHisham121/AI--powered-Chatbots.git](https://github.com/MalakHisham121/AI--powered-Chatbots.git)
cd AI--powered-Chatbots

2. Install Dependencies

Install the required packages (including LlamaIndex, LangGraph, and Neo4j drivers) from the requirements file:

pip install -r requirements.txt

3. Environment Variables

Create a .env file in the root directory and add your database and LLM credentials:

# LLM Provider
GROQ_API_KEY=your_groq_api_key_here

# Neo4j Database Configuration
NEO4J_URI=neo4j+s://your-instance.databases.neo4j.io
NEO4J_USER=neo4j
NEO4J_PASSWORD=your_neo4j_password
NEO4J_DATABASE=neo4j

💻 Usage Instructions

Running the Inventory Chatbot (SQL)

First, initialize the SQLite database with the sample schema:

cd "AI inventory Chatbot"
python db-setup.py

Then, start the interactive terminal agent:

python main.py

Running the Knowledge Graph Agent (Neo4j)

Ensure your Neo4j instance is running, then execute the agent:

cd "AI Knowledge Graph Agent"
python main.py

Example Graph Commands:

  • Add: "Cairo University is located in Egypt."
  • Inquire: "Where is Cairo University located?"
  • Edit: "Update the located_in relation for Cairo University to Giza."
  • Delete: "Remove the located_in relation from Cairo University."

📓 Bonus: SQL to NL API Notebook

The repository also includes AI_Agent_to_convert_from_SQL_to_NL.ipynb, an experimental Jupyter Notebook. It demonstrates how to wrap a Gemini 1.5 Pro model in a FastAPI endpoint to generate SQL, execute it against a local SQLite database, and return a natural language explanation alongside performance metrics (latency) in a clean JSON response.