Module 10 Wrap-up: Your Private Library Bot

You have learned the theory of vectors, the mechanics of chunking, and the architecture of pipelines. Now, it’s time to build a real-world RAG system. We are going to build a Python script that reads a "Knowledge" folder and answers questions based on it.

---

Hands-on Exercise: The Knowledge Bot

1. Requirements

Ensure you have the necessary libraries: pip install ollama chromadb

2. The Project Setup

Create a folder named knowledge and put a few .txt files inside it (e.g., your favorite recipes, a summary of a book, or your resume).

3. The Code

Create a file named rag_bot.py:

import ollama
import chromadb

# 1. Initialize DB and Ollama
client = chromadb.Client()
collection = client.create_collection(name="docs")

# 2. Ingest your data (Simulated for this lesson)
docs = [
    "The secret ingredient in the sauce is honey and ginger.",
    "The office is closed on the first Monday of every month.",
    "The project deadline has been moved to December 15th."
]

for i, d in enumerate(docs):
  response = ollama.embeddings(model="mxbai-embed-large", prompt=d)
  collection.add(
    ids=[str(i)],
    embeddings=[response["embedding"]],
    documents=[d]
  )

# 3. The Search Function
query = "What is in the secret sauce?"
query_embed = ollama.embeddings(model="mxbai-embed-large", prompt=query)["embedding"]

results = collection.query(query_embeddings=[query_embed], n_results=1)
context = results["documents"][0][0]

# 4. The Answer
output = ollama.generate(
  model="llama3",
  prompt=f"Answer the question based ONLY on this context: {context}\n\nQuestion: {query}"
)

print(f"Question: {query}")
print(f"Answer: {output['response']}")

---

Module 10 Summary

• RAG gives LLMs access to specific, private, or current data.
• Embeddings turn text into a "Searchable Map."
• Vector Databases like ChromaDB store these maps locally.
• Chunking ensures the AI can find specific facts inside long documents.

Coming Up Next...

In Module 11, we move to the final frontier of customization: Fine-Tuning. We will learn when you should stop using RAG and start training the model's weights directly using Adapters and LoRA.

---

Module 10 Checklist

• I can explain what a "Vector" is to a non-technical person.
• I have downloaded the mxbai-embed-large model.
• I understand why "Overlap" is important for chunking.
• I have successfully run a "Similarity Search" in a Vector Store.
• I can describe the 3 steps of a RAG pipeline (Retrieve, Augment, Generate).