RAG Support Assistant: Microsoft Products

Business Problem

Customer support teams at Microsoft handle thousands of questions daily across Teams, Azure, Copilot, Surface, Windows, Office 365, and GitHub. Manually answering repetitive questions is expensive and slow. This project builds an AI-powered support assistant using Retrieval Augmented Generation (RAG) that automatically answers customer questions by retrieving relevant documentation and generating accurate, grounded responses using GPT-4o.

What is RAG?

RAG (Retrieval Augmented Generation) is the architecture that powers Microsoft Copilot. It works in three steps:

Customer Question → Retrieve Relevant Docs → GPT-4o Generates Answer

Unlike a standard LLM, a RAG system only answers based on retrieved context, preventing hallucinations and ensuring factually grounded responses.

Architecture

Knowledge Base (10 docs)
        ↓
TF-IDF Vectorizer (similarity search)
        ↓
Retrieved Documents (top 3 most relevant)
        ↓
GPT-4o (Azure OpenAI) generates grounded answer
        ↓
Response with sources and relevance scores

Products Covered

Product	Documents
Microsoft Teams	2
Microsoft Copilot	2
Azure	2
Surface	1
Windows	1
Office 365	1
GitHub	1

Key Features

• Grounded responses: Answers based only on retrieved documents, never invented
• Source citations: Every answer cites which documents were used and their relevance scores
• Hallucination prevention: Out of scope questions are handled honestly without making up answers
• Multi-turn conversations: Maintains context across follow up questions in the same session
• Relevance scoring: Shows confidence scores for retrieved documents

Sample Interaction

Question: My Teams keeps crashing during important meetings

Answer: If Microsoft Teams keeps crashing during your meetings, follow these steps:

1. Update Teams via Help > Check for updates
2. Clear the Teams cache at %appdata%/Microsoft/Teams
3. Check internet connection, Teams requires 1.5 Mbps for HD video
4. Disable GPU hardware acceleration in Settings > General
5. Reinstall Teams if the issue persists

Sources used:

• Microsoft Teams: Teams crashes during meetings (relevance: 40.80%)

Hallucination Prevention Results

Question	In Knowledge Base	Response
Teams crashing fix	Yes	Accurate grounded answer
Azure bill too high	Yes	Accurate grounded answer
Price of Teams	No	Correctly declined to answer
Python + Azure SQL	No	Correctly declined to answer
Windows 12 requirements	No	Correctly declined to answer

Tech Stack

• Python 3.9
• Azure OpenAI (GPT-4o)
• scikit-learn (TF-IDF retrieval)
• python-dotenv (secure credential management)
• openai Python SDK

Project Structure

rag-support-assistant/
├── notebooks/
│   └── 01_rag_support_assistant.ipynb   # Full RAG pipeline
├── src/
│   └── (helper scripts)
├── .env                                  # API credentials (not pushed)
├── .gitignore
├── README.md
└── requirements.txt

How to Run

# Clone the repo
git clone https://github.com/philipatosam/rag-support-assistant.git
cd rag-support-assistant

# Create and activate virtual environment
python3 -m venv venv
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Create .env file with your Azure OpenAI credentials
echo "AZURE_OPENAI_KEY=your-key-here" > .env
echo "AZURE_OPENAI_ENDPOINT=your-endpoint-here" >> .env
echo "AZURE_OPENAI_DEPLOYMENT=gpt-4o" >> .env

# Launch Jupyter
jupyter notebook

Azure OpenAI Setup

1. Create an Azure account at https://azure.microsoft.com/free
2. Create an Azure OpenAI resource in the Azure Portal
3. Deploy a GPT-4o model in Azure AI Foundry
4. Copy your endpoint and API key to the .env file

What I Would Do Next

• Replace TF-IDF retrieval with Azure AI Search for semantic search
• Expand knowledge base with real Microsoft documentation
• Add conversation memory using Azure Cosmos DB
• Build a web interface using Streamlit or FastAPI
• Implement response evaluation using GPT-4o as a judge
• Add support ticket auto-classification and routing