RAG and Function Calling Demo

A simple educational project showing how RAG (Retrieval Augmented Generation) and Function Calling work with LLMs.

What This Demonstrates

RAG (Retrieval Augmented Generation)

• How to convert documents into vectors (embeddings)
• How to find relevant documents using similarity search
• How to use retrieved documents as context for better answers

Function Calling

• How LLMs can call external tools/functions
• Simple examples: time lookup and calculations
• How to route queries to appropriate functions

Quick Start

1. Install Ollama:

   # Visit https://ollama.ai and install
   ollama serve
   ollama pull llama3.2:3b

2. Install dependencies:

   pip install -r requirements.txt

3. Run the demo:

   python chatbot.py

Demo Flow

The demo shows both concepts in action:

Function Calling Demo

User: What time is it?
Bot: The current time is 2024-01-15 14:30:25

User: Calculate 15 * 7 + 23
Bot: 15 * 7 + 23 = 128

RAG Demo

Adding documents: "Python is a programming language...", "Machine learning..."

User: Who created Python?
Bot: Based on the documents, Python was created by Guido van Rossum in 1991.

Code Structure

• src/main.py - Simple demo runner
• src/function_agent.py - Function calling logic (50 lines)
• src/rag_agent.py - RAG implementation (100 lines)
• test_demo.py - Basic tests

Key Concepts Shown

RAG Process:

1. Convert documents to embeddings (vectors)
2. Convert user query to embedding
3. Find most similar documents (cosine similarity)
4. Use relevant documents as context in LLM prompt

Function Calling Process:

1. Analyze user query for function needs
2. Extract parameters (e.g., math expression)
3. Call appropriate function
4. Format result into natural response

Educational Value

This project strips away complexity to show the core concepts:

• No complex frameworks (minimal dependencies)
• Clear, readable code with comments
• Step-by-step process demonstration
• Easy to modify and experiment with

Requirements

• Python 3.7+
• Ollama running locally
• Dependencies: ollama, sentence-transformers, numpy

Perfect for learning how modern AI applications work under the hood!