This project is a Retrieval-Augmented Generation (RAG) chatbot for network engineering documentation.

• ⚡ Runs entirely on CPU – no GPU required (This makes it lightweight and portable, but responses may be slower compared to GPU setups.)
• 🔒 Privacy-focused – everything is processed locally on your machine:
  • Documents are never uploaded to the cloud
  • Embeddings are stored in a local Chroma database
  • LLM inference is done through Ollama, fully offline

This makes the chatbot lightweight, portable, and secure for use with sensitive networking documentation.

• Ollama for running local LLMs (e.g., llama3.2:3b)
• Chroma as the vector database
• HuggingFace embeddings for semantic search
• Unstructured for PDF/Doc parsing
• LangChain Text Splitters for clean chunking
• Rich for colored terminal UI

The chatbot retrieves context from your networking documentation (PDFs) and answers queries with proper citations.

. ├── .env # Environment variables ├── chat.py # Interactive chatbot loop ├── RAG_network_documents.py # Script to process docs & build vector DB ├── tools.py # Utilities (embedding, retriever, logging, checks) ├── docs/ # Place your PDF docs here │ ├── CISCO - XR │ │ └── Cisco XR 12000 Series Router.pdf │ │ │ ├── HUAWEI - NE40e │ │ └── NE40E Product Description.pdf │ │ │ └── NOKIA - SR-7750 │ └── nokia-7750-service-router-datasheet-en.pdf │ ├── chromadb/ # Chroma persistence directory (auto-created) ├── chat_logs/ # Saved chat logs (JSON + Markdown, auto-created) 

pip install -r requirements.txt

Start the Ollama server:

ollama serve

Edit the .env file:

MODEL_NAME="llama3.2:3b" CHROMA_DB_PATH="./chromadb" COLLECTION_NAME="network_docs" DOCS_PATH="./docs" EMBEDDING_MODEL_NAME="intfloat/multilingual-e5-large-instruct" TOP_K=5

To process your networking PDFs into embeddings and store them in Chroma:

python RAG_network_documents.py

This will:

1. Parse PDFs from ./docs using unstructured
2. Chunk the text using LangChain text splitters
3. Generate embeddings
4. Store them in the Chroma DB

Start an interactive chat:

python chat.py

• Colored terminal UI
• Context-aware answers with citations
• /exit, /quit, /q to exit
• Empty queries are ignored
• Chat logs saved as JSON + Markdown on exit

This project was inspired by and adapted from: Ray End-to-End RAG Tutorial

LinkedIn: Lahcen Khouchane WebSite : NetOpsAutomation.com