📘 Financial Inventory RAG Chatbot with Neo4j Knowledge Graph

This application enables Retrieval-Augmented Generation (RAG) on a complex Excel dataset containing multi-level inventory data across different group companies. It uses Neo4j Knowledge Graph for structured representation and LangChain with Azure OpenAI for intelligent querying and answer generation.

---

💻 Python Environment Setup

1. Create a virtual environment:

python -m venv code-rag-env
source code-rag-env/bin/activate  # or .\code-rag-env\Scripts\activate on Windows

2. Install dependencies:

pip install -r requirements.txt

---

📌 Features

• Converts Excel-based multi-level inventory data into a Neo4j Knowledge Graph
• Performs semantic RAG-style Question Answering over the graph
• Uses LangChain + Azure OpenAI to convert natural language to Cypher and generate accurate responses
• Streamlit-powered UI for easy interaction

---

🧩 Data Schema

• Inventory Levels: InventoryLevel1 -> InventoryLevel2 -> InventoryLevel3 -> InventoryLevel4

• Relationships:

  • (:InventoryLevelX)-[:SUB_INVENTORY]->(:InventoryLevelY)
  • (:GroupCompany)-[:HAS_VALUE {amount, year}]->(:InventoryLevel4)

---

⚙️ Setup Instructions

🔹 Step 1: Neo4j Desktop Setup

1. Download and install Neo4j Desktop

2. Create a new Project and inside it, a new Local DBMS

3. Start the database and get:

   • Bolt URL (e.g., bolt://localhost:7687)
   • Username (default: neo4j)
   • Password (set by you)

4. Create a .env file in the root directory:

NEO4J_URI=bolt://localhost:7687
NEO4J_USER=neo4j
NEO4J_PASSWORD=your_password_here
AZURE_DEPLOYMENT=your_azure_deployment_name
AZURE_OPENAI_API_KEY=your_openai_key
AZURE_OPENAI_ENDPOINT=https://your-resource.openai.azure.com/
AZURE_OPENAI_VERSION=your-api-version

🔹 Step 2: Create the Knowledge Graph

📝 Update the .env file with the correct Neo4j and Azure OpenAI credentials before executing this step.

Use the notebook create_kg.ipynb to:

• Load Excel file (containing inventory and company data)

• Extract and process:

  • Inventory Levels (L1 to L4)
  • Relationships between inventories
  • GroupCompany to Inventory mappings with properties amount and year

• Push structured nodes and relationships to Neo4j

🔹 Step 3: RAG Pipeline Execution

The core RAG logic is defined in main.py:

1. Normalization Layer

   • Uses LLM to extract structured fields: company_name, inventory_type, year, and canonicalized_query

2. Cypher Generation Layer

   • Generates a Cypher query from the canonicalized query using LangChain prompts

3. Query Execution

   • Runs the Cypher on Neo4j and extracts the relevant data

4. Answer Generation Layer

   • Uses LLM to synthesize an answer from Cypher results + user query

🔹 Step 4: Chatbot UI

Run the Streamlit app with:

streamlit run streamlit_app.py

• Ask financial questions like:

  • "Which inventory tools has the highest difference for AS from 2023 to 2024?"
  • "Show amount change for Machinery across years"

• View normalized query, generated Cypher, and final answer

---

📁 File Structure

├── create_kg.ipynb         # Creates and loads Neo4j KG from Excel
├── main.py                 # Core RAG pipeline: query normalization, Cypher gen, LLM response
├── prompts.py              # Required Prompts for RAG and QA 
├── streamlit_app.py        # Streamlit chatbot interface
├── .env                    # Neo4j and Azure OpenAI secrets
├── requirements.txt        # Required packages

---

🧠 Tech Stack

• Neo4j – Graph DBMS for structured inventory data
• LangChain – RAG pipeline orchestration
• Azure OpenAI – LLM backend (ChatGPT)
• Streamlit – UI for chatbot
• Python – Core implementation

⚠️ Note: The Excel data file is not provided in this repository for security and confidentiality reasons.