Model Anatomy · Episode 06 · 1:38
RAG explained from scratch: how an AI answers from your own documents
Split the docs into chunks, turn them into vectors and find the closest match to the question. "How many vacation days do new hires get?" The best chunk scores 0.91 and the answer, 15 days, comes with its source.
What you’ll needPython 3.10+
What you’ll need
- python.org/downloads →
Python 3.10 or newerRuns the code. Check yours with python3 --version. - code.visualstudio.com →
VS Code or CursorEither works; so does any editor you like. - git-scm.com/downloads →
gitGets the code from GitHub. - No packages needed: it uses only the Python standard library.
Set it up
- 1Get the code
git clone https://github.com/DayanEbrar0X/data-anatomy.ai.git cd data-anatomy.ai
- 2Make a virtual environmentIt keeps this project's packages separate from the rest of your computer.
python3 -m venv .venv source .venv/bin/activate
- 3Run it
cd ai-engineering/06-rag python3 src/rag.py
Project structure06-rag/
Project structure
ai-engineering/06-rag/Open on GitHub →
- data/
- handbook.json8 chunks of HR, IT and finance policy
- short/
- docs.pyloads the chunks, builds the vocabulary
- rag.pyrunthe code from the video
- src/
- docs.pyloads the chunks, builds the vocabulary
- rag.pymainthe code from the video
- README.mdthe idea, how to run it, a line-by-line walkthrough, things to try
ai-engineering/06-rag/src/rag.pyOpen on GitHub →
import math, re
from docs import DOCS, VOCAB
def embed(text):
words = re.findall(r"[a-z]+", text.lower())
return [words.count(w) for w in VOCAB]
def cosine(a, b):
dot = sum(x * y for x, y in zip(a, b))
return dot / (math.hypot(*a) * math.hypot(*b))
ask = "How many vacation days do new hires get?"
q = embed(ask)
scores = {d: cosine(q, embed(t))
for d, t in DOCS.items()}
top = sorted(scores, key=scores.get,
reverse=True)[:2]
context = "\n".join(DOCS[d] for d in top)
prompt = f"{context}\n\nQuestion: {ask}"
for d in top:
print(d, f"{scores[d]:.2f}", DOCS[d])
print("sources:", *top)Transcript
Your company's AI never read your HR handbook. Ask about vacation days, and it guesses. The fix: look it up first. That's RAG: retrieval-augmented generation.
Like an open-book exam: find the right page, then answer. An employee asks: how many vacation days do new hires get? First, cut the handbook into eight chunks. Embed turns a chunk into numbers, called a vector.
It counts each vocabulary word in the text. Real systems use a learned model here. Every chunk is now a point. Similar text lands close.
Cosine asks: do two vectors point the same way? Multiply matching counts, divide by length. One is a perfect match. The question is embedded the same way.
We score every chunk against it, then keep the top two. Their text goes into the prompt, above the question. Then print what came back. Let's run it.
Top match: hr-1, 0.91. New hires get 15 vacation days. Second, the laptop rule: same words, different meaning.
A model reading that prompt can answer: 15 days, citing hr-1. Now swap these chunks for your wiki pages, tickets and contracts. Edit a doc, and the answer changes. No retraining.
Every answer cites a source, and you control what gets retrieved. Find the right page, then answer. That's RAG.
Read the lesson on GitHub →- #rag
- #llm
- #python
- #genai
