Making AI accessing your personal work notes using RAG mechanism to be your work assistant

Have you ever wished ChatGPT could answer questions about YOUR personal notes? Like “What are my lab credentials?” or “Who do I contact for hardware purchases?” – things only you would know from your private documentation?

In this article, I’ll show you how to build a simple RAG (Retrieval-Augmented Generation) system that lets you chat with OpenAI while it seamlessly pulls context from your local Obsidian notes when needed.

NOTE#1 : I was historically using OneNote, but migrating to Obsidian is something you can do using some export/import tools. So this project was my personal motivation to that 🙂

NOTE#2 : This article uses openai example, but when you download my code from github, you will see in the config.py file, that you can actually point it to a local AI running on LM Studio. This is the way forward if your work notes contain sensitive information you do not want OpenAI to learn about as using public API exposes your work notes to Sam Altman and their cronies to train their AI models on.

What We’re Building

We’re creating an AI assistant that:

Chats normally with you using OpenAI’s GPT models
Automatically detects when a question might relate to your personal notes
Retrieves relevant information from your local knowledge base
Combines AI intelligence with YOUR specific documentation

The best part? Your notes stay local – only relevant snippets are sent to the AI when needed.

Getting Started

Step 1: Clone the Repository

git clone https://github.com/zerxen/AI_NetworkGeekStuff_Assistant
cd AI_NetworkGeekStuff_Assistant

Step 2: Install Dependencies

pip install -r requirements.txt

Step 3: Configure Your API Key

Copy the template and add your OpenAI API key:

cp config.py_template config.py

Edit config.py and replace the placeholder with your actual key:

OPENAI_API_KEY = "sk-your-actual-api-key-here"
MODEL = "gpt-4o-mini"

You can get an API key from OpenAI Platform. Note that API usage requires a paid account (minimum $5 credit).

Understanding the Knowledge Sources

The project includes example knowledge sources in the knowledge_sources/ directory. Let’s look at what’s there:

Example 1: Company Contacts (acmeco/contacts/)

# ACME Co Internal Contacts

## Procurement Department

| Name | Role | Email | Phone | Notes |
|------|------|-------|-------|-------|
| Sarah Mitchell | Procurement Manager | s.mitchell@acmeco.com | ext. 2401 | Approves orders over $5000 |
| Dave Chen | Hardware Buyer | d.chen@acmeco.com | ext. 2405 | Contact for network equipment |

## IT / Network Team (Colleagues)

| Name | Role | Email | Phone | Notes |
|------|------|-------|-------|-------|
| Tom Bradley | Network Architect | t.bradley@acmeco.com | ext. 4501 | Senior, 10+ years |
| Rachel Kim | Security Analyst | r.kim@acmeco.com | ext. 4520 | Firewall changes need her approval |

Example 2: Lab Credentials (acmeco/lab_documentation/)

# Lab Device Credentials

> Note: These are LAB credentials only.

## Network Devices (Cisco)

**Standard Lab Login**:
- Username: `labadmin`
- Password: `LabTest123!`
- Enable: `EnableLab!`

**Applies to**: Lab-FW-01, Lab-SW-01, Lab-SW-02, R1, R2, R3

Example 3: Personal Recipes (Recipes/)

Yes, even your cooking notes work! The system indexes everything in markdown format:

**Quick Omelette Recipe (1 serving)**

**Ingredients**
- 2-3 eggs
- Salt and black pepper
- 1 tbsp butter or oil

**Steps**
1. Crack the eggs into a bowl, add a pinch of salt and pepper...

Testing the RAG System

Before chatting, let’s verify the RAG pipeline works. The project includes test_rag.py for this purpose.

Running the Test

python test_rag.py

You’ll see output like this:

============================================================
Testing RAG Pipeline Initialization
============================================================

1. Initializing RAG Manager...
RAG Manager created (vector store will be initialized on first query)
   Clearing database as requested...
Vector store cleared

2. Testing document retrieval...
   Query 1: 'cisco lab credentials'
   Retrieved relevant documents
   Preview: ## Relevant Knowledge from Sources:

**[1] From lab_credentials.md:**
# Lab Device Credentials > Note: These are LAB credentials only...

Customizing Test Queries

Want to test different queries? Edit line 38 in test_rag.py:

test_queries = [
    "cisco lab credentials",
    "who handles procurement",
    "omelette recipe"
]

Each query will search your knowledge base and return relevant matches.

How Vectorization Works (The Magic Behind RAG)

Here’s where it gets interesting. How does the system know that “cisco lab credentials” should return your credentials file?

Text to Numbers

When you run the system, your markdown files are converted into embedding vectors – arrays of numbers that represent the semantic meaning of the text. OpenAI’s text-embedding-3-small model does this conversion.

For example, conceptually:

"Lab Device Credentials - Cisco - Username: labadmin"
    -> [0.023, -0.156, 0.891, 0.044, ... ] (1536 dimensions)

"Network Devices - Password: LabTest123!"
    -> [0.019, -0.148, 0.887, 0.051, ... ] (similar vector)

"Quick Omelette Recipe - eggs, butter"
    -> [-0.234, 0.567, 0.123, -0.445, ... ] (very different vector)

Similarity Search

When you query “cisco lab credentials”:

Your query is converted to a vector: [0.021, -0.152, 0.889, ...]
ChromaDB finds vectors that are mathematically “close” to yours
The credentials document has a similar vector -> it’s returned as relevant
The recipe has a very different vector -> it’s ignored

This is called cosine similarity – measuring the angle between vectors. Similar meaning = similar direction = small angle = high similarity.

Local Storage

All embeddings are stored locally in the chroma_db/ folder. Your notes never leave your machine until you actually ask a question that needs them.

Chatting with the AI Assistant

Now for the main event – let’s chat with the AI while it uses your knowledge base.

Starting the Chat

python main.py

Initializing RAG knowledge base...
Knowledge base ready.

Interactive ChatGPT CLI (type 'exit' or Ctrl-C to quit)
You:

Example Conversation

You: What are the lab credentials for Cisco devices?

The AI recognizes this might need local knowledge and calls the retrieveKnowledge tool:

Tool executed called: retrieveKnowledge with query='cisco lab credentials' top_k=5
.. results of tool call provided to the context

ChatGPT (after tools): Based on your lab documentation, the Cisco device
credentials are:

- Username: labadmin
- Password: LabTest123!
- Enable password: EnableLab!

These apply to Lab-FW-01, Lab-SW-01, Lab-SW-02, R1, R2, and R3.

You: Who should I contact about buying new network switches?

ChatGPT (after tools): For network equipment purchases, you should contact
Dave Chen (Hardware Buyer) at d.chen@acmeco.com, extension 2405.

For orders over $5000, you'll also need approval from Sarah Mitchell
(Procurement Manager) at extension 2401.

You: How do I make an omelette?

ChatGPT (after tools): Here's the quick omelette recipe from your notes:

1. Crack 2-3 eggs into a bowl, add salt and pepper, whisk briefly
2. Heat butter in a non-stick pan over medium heat
3. Pour in eggs, swirl to spread evenly
4. Add fillings to one half when edges set (30-60 seconds)
5. Fold in half, cook another 30-60 seconds
6. Serve immediately!

How It Decides When to Use RAG

The magic happens in the system prompt. The AI is instructed:

“Use the ‘retrieveKnowledge’ tool whenever you need to reference specific documentation, technical details, or look up configuration examples. Only use it when necessary – avoid overusing it for general knowledge.” So if you ask “What is BGP?” – it answers from general knowledge. But “What’s MY lab’s BGP configuration?” triggers a knowledge retrieval.

Setting Up Your Own Knowledge Base

Replace the example files with your actual notes:

Delete the example folders in knowledge_sources/
Copy your Obsidian vault (or any markdown files) into knowledge_sources/
Run python test_rag.py to rebuild the database

The system will automatically:

Scan all .md files recursively
Split them into intelligent chunks (respecting markdown structure)
Generate embeddings and store them locally

A Personal Note

The example data in this project (the “acmeco” company) is actually a sanitized version of my real Obsidian vault. I’ve been using this system to help me:

Find old meeting notes – “What did we discuss in the January network planning meeting?”
Look up contacts – “Who handles software licensing again?”
Remember lab details – “What’s the management IP for the firewall?”

After years of accumulating notes, it became impossible to remember where everything was. Now I just ask the AI, and it finds the relevant note for me – like having a personal assistant who has read everything I’ve ever written.

Conclusion

RAG bridges the gap between powerful AI models and your personal knowledge. The AI provides intelligence and natural conversation; your notes provide the specific, private information only you have.

The complete code is available at: https://github.com/zerxen/AI_NetworkGeekStuff_Assistant

Give it a try with your own Obsidian vault – you might be surprised how useful it is to finally have an AI that actually knows YOUR stuff.

Have questions or improvements? Open an issue on GitHub or drop a comment below!

kokotina