RAG Systems Explained: A Complete Beginner's Guide for 2025

┌─────────────────────────────────────────────────────────────┐
│                    RAG SYSTEM PIPELINE                      │
└─────────────────────────────────────────────────────────────┘

User Question: "What is our company's remote work policy?"
                                ↓
┌─────────────────────────────────────────────────────────────┐
│  STEP 1: RETRIEVE (Find Relevant Information)              │
├─────────────────────────────────────────────────────────────┤
│  1. Convert question to embedding (vector)                  │
│  2. Search vector database for similar embeddings           │
│  3. Retrieve top 3-5 most relevant document chunks          │
│                                                             │
│  Retrieved Chunks:                                          │
│  → "Remote employees are entitled to flexible hours..."    │
│  → "All employees must use VPN when accessing..."          │
│  → "Work-from-home equipment reimbursement up to $500..."  │
└─────────────────────────────────────────────────────────────┘
                                ↓
┌─────────────────────────────────────────────────────────────┐
│  STEP 2: AUGMENT (Combine Question + Retrieved Docs)       │
├─────────────────────────────────────────────────────────────┤
│  Build prompt with:                                         │
│  - Original user question                                   │
│  - Retrieved document chunks                                │
│  - Instructions (answer using only provided context)        │
└─────────────────────────────────────────────────────────────┘
                                ↓
┌─────────────────────────────────────────────────────────────┐
│  STEP 3: GENERATE (AI Creates Answer from Context)         │
├─────────────────────────────────────────────────────────────┤
│  LLM receives augmented prompt and generates response       │
│                                                             │
│  Response:                                                  │
│  "According to our company policy, remote employees        │
│   have flexible hours and are eligible for up to $500      │
│   in equipment reimbursement. All remote access requires   │
│   VPN connection for security."                            │
└─────────────────────────────────────────────────────────────┘

You are a helpful assistant. Answer the user's question using
ONLY the information provided in the context below. If the answer
is not in the context, say "I don't have that information."

CONTEXT:
---
Document 1: Remote employees are entitled to flexible work
hours between 6 AM and 10 PM in their local timezone...

Document 2: All employees working remotely must connect via
the company VPN when accessing internal systems...

Document 3: Work-from-home equipment reimbursement is available
up to $500 per year for items such as desks, chairs, monitors...
---

USER QUESTION: What is our company's remote work policy?

ANSWER:

# Install required libraries
pip install openai chromadb tiktoken

# You'll need an OpenAI API key
# Get one at: https://platform.openai.com/api-keys

# sample_documents.py

documents = [
    {
        "id": "doc1",
        "text": "Remote employees are entitled to flexible work hours between 6 AM and 10 PM in their local timezone. Core meeting hours are 10 AM to 3 PM EST.",
        "metadata": {"source": "employee_handbook", "section": "remote_work"}
    },
    {
        "id": "doc2",
        "text": "All employees working remotely must connect via the company VPN when accessing internal systems, databases, or customer data. VPN credentials are issued by IT.",
        "metadata": {"source": "security_policy", "section": "vpn"}
    },
    {
        "id": "doc3",
        "text": "Work-from-home equipment reimbursement is available up to $500 per year for items such as desks, chairs, monitors, keyboards, and ergonomic accessories.",
        "metadata": {"source": "benefits_guide", "section": "equipment"}
    },
    {
        "id": "doc4",
        "text": "Remote employees must attend mandatory all-hands meetings via video conference on the first Monday of each month at 2 PM EST.",
        "metadata": {"source": "employee_handbook", "section": "meetings"}
    }
]

# build_rag_index.py

import openai
import chromadb
from chromadb.config import Settings

# Initialize OpenAI
openai.api_key = "your-api-key-here"

# Initialize Chroma vector database
chroma_client = chromadb.Client(Settings(
    chroma_db_impl="duckdb+parquet",
    persist_directory="./chroma_db"
))

# Create or get collection
collection = chroma_client.get_or_create_collection(
    name="company_docs",
    metadata={"description": "Company documentation for RAG"}
)

# Generate embeddings and add to database
def embed_and_store(documents):
    for doc in documents:
        # Generate embedding using OpenAI
        response = openai.embeddings.create(
            model="text-embedding-3-small",
            input=doc["text"]
        )
        embedding = response.data[0].embedding

        # Store in Chroma
        collection.add(
            ids=[doc["id"]],
            embeddings=[embedding],
            documents=[doc["text"]],
            metadatas=[doc["metadata"]]
        )
        print(f"Added {doc['id']} to vector database")

# Run indexing
from sample_documents import documents
embed_and_store(documents)
print("✅ Indexing complete!")

# rag_query.py

import openai
import chromadb
from chromadb.config import Settings

openai.api_key = "your-api-key-here"

# Connect to existing Chroma database
chroma_client = chromadb.Client(Settings(
    chroma_db_impl="duckdb+parquet",
    persist_directory="./chroma_db"
))
collection = chroma_client.get_collection(name="company_docs")

def rag_query(user_question, top_k=3):
    """
    RAG pipeline: Retrieve → Augment → Generate
    """
    # STEP 1: RETRIEVE
    # Generate embedding for user question
    question_embedding = openai.embeddings.create(
        model="text-embedding-3-small",
        input=user_question
    ).data[0].embedding

    # Query vector database for similar documents
    results = collection.query(
        query_embeddings=[question_embedding],
        n_results=top_k
    )

    retrieved_docs = results['documents'][0]

    # STEP 2: AUGMENT
    # Build context from retrieved documents
    context = "\n\n".join([f"Document {i+1}: {doc}"
                            for i, doc in enumerate(retrieved_docs)])

    # Build augmented prompt
    prompt = f"""You are a helpful assistant. Answer the user's question using ONLY the information provided in the context below. If the answer is not in the context, say "I don't have that information."

CONTEXT:
---
{context}
---

USER QUESTION: {user_question}

ANSWER:"""

    # STEP 3: GENERATE
    # Send to LLM for answer generation
    response = openai.chat.completions.create(
        model="gpt-4",
        messages=[
            {"role": "system", "content": "You are a helpful company assistant."},
            {"role": "user", "content": prompt}
        ],
        temperature=0
    )

    answer = response.choices[0].message.content

    return {
        "question": user_question,
        "answer": answer,
        "sources": retrieved_docs
    }

# Test the RAG system
if __name__ == "__main__":
    question = "What is our company's remote work policy?"
    result = rag_query(question)

    print(f"\n❓ QUESTION: {result['question']}")
    print(f"\n✅ ANSWER: {result['answer']}")
    print(f"\n📚 SOURCES USED:")
    for i, source in enumerate(result['sources'], 1):
        print(f"  {i}. {source[:100]}...")

# Run the RAG query
python rag_query.py

# Expected output:
❓ QUESTION: What is our company's remote work policy?

✅ ANSWER: According to our company policy, remote employees have flexible work hours between 6 AM and 10 PM in their local timezone, with core meeting hours from 10 AM to 3 PM EST. All remote workers must connect via company VPN when accessing internal systems. Additionally, work-from-home equipment reimbursement is available up to $500 per year for items like desks, chairs, and monitors.

📚 SOURCES USED:
  1. Remote employees are entitled to flexible work hours between 6 AM and 10 PM...
  2. All employees working remotely must connect via the company VPN...
  3. Work-from-home equipment reimbursement is available up to $500 per year...

# Query with metadata filtering
results = collection.query(
    query_embeddings=[question_embedding],
    n_results=5,
    where={
        "source": "employee_handbook",  # Only retrieve from handbook
        "last_updated": {"$gt": "2024-01-01"}  # Only recent docs
    }
)

from cohere import Client

# Initialize Cohere for re-ranking
co = Client(api_key="your-cohere-key")

# Re-rank retrieved documents
reranked = co.rerank(
    query=user_question,
    documents=retrieved_docs,
    top_n=3,
    model="rerank-english-v2.0"
)

# Use top 3 re-ranked docs in prompt
final_docs = [doc.document for doc in reranked.results]

# Original query: "remote work policy"

# Expanded queries:
queries = [
    "remote work policy",
    "work from home guidelines",
    "telecommuting rules and requirements",
    "distributed team work arrangements"
]

# Retrieve for each query, combine and de-duplicate results

# User asks: "What was our Q4 2024 revenue?"

# LLM analyzes question and generates:
{
    "collection": "financial_reports",
    "filters": {
        "year": 2024,
        "quarter": "Q4",
        "type": "revenue"
    },
    "search_query": "quarterly revenue figures"
}