Interactive Learning Lab

RAG Fundamentals

Two classes, six labs. First the engine — how text becomes numbers, how numbers carry meaning, how a vector database searches them. Then the pipeline built on top of it.

text-embedding-3-small GPT-4o-mini Real API calls Nothing hidden
Class 4A — Semantic Search: The Engine
Why search by meaning is needed, how text becomes numbers, how the numbers carry meaning, and how a vector database scales it.
01
🤥
The Problem
The Hallucination Problem
Ask GPT-4o-mini about NovaCorp's private HR policy and watch it confidently invent answers.
→
02
🔍
The Problem
Keyword vs Semantic Search
Search the same corpus two ways. See where exact keyword matching fails and search by meaning succeeds.
→
03
🔢
The Engine
Embeddings: Text Becomes Numbers
Any text → 1,536 numbers. Compare sentences, read the score correctly — scores are relative, rank them.
→
04
🗄️
The Engine
Retrieval by Hand & the Vector Store
Five sentences, one question: score, sort, keep the top 3. Then store them, add a new version, delete the old one.
→
Class 4B — The RAG Pipeline
Load → Chunk → Embed → Index → Retrieve → Augment → Generate, on a real document.
05
✂️
The Pipeline
Indexing a Document
Upload a PDF and step through extract → chunk → embed → index → search.
→
06
🚀
The Pipeline
The Full RAG Pipeline
Question → embed → retrieve → build the prompt → grounded answer, every step visible.
→
The Story Arc
🤥
Labs 1–2
The requirement: hallucination, and search that matches spelling not meaning
→
🔢
Labs 3–4
The engine: embeddings, similarity, retrieval, the vector store
→
🚀
Labs 5–6
The pipeline: index a document, answer from it