// CODE PLAYGROUND

Hands-On Lab Sandbox

Every lab from the curriculum, in three modes: walk through the code section by section, practice tasks with AI verification, or experiment freely with the Gemini explainer.

// HANDS_ON_LAB

Build a naive RAG from scratch

python3 tasks

// WALKTHROUGH1 / 5 sections revealed

1. Set up the corpus

RAG starts with documents. In real systems they come from PDFs, web pages, or a database; here we hard-code four short strings to keep the focus on the pipeline. Each string is one 'chunk'. In production, long documents get split into 200–800-token chunks first — but the rest of the pipeline doesn't change.

// KEY POINTS

One chunk per array entry. Real systems split long text first.
The OpenAI client reads `OPENAI_API_KEY` from the environment.

// WHAT TO NOTICE

These four chunks are intentionally about different subtopics. That makes retrieval trivial here — in production, similar chunks are the hard case.

snippet

from openai import OpenAI
import numpy as np

client = OpenAI()
DOCS = [
    "Pinecone is a managed vector database...",
    "Weaviate supports hybrid (BM25 + vector) search...",
    "BGE-M3 is an open-source multilingual embedding model...",
    "RAGAS measures faithfulness, answer relevancy, context precision/recall.",
]

// LOADING