Retrieval · chunking

Chop it up and rank the pieces

🎓 School maths · Years 9–10Number — ranking & comparison — Split text into pieces, score each piece for relevance and order them.Maths by year →

The Case. Real documents are long — a whole page, not one line. AI splits them into chunks first, then ranks every chunk against your question to find the one that actually answers it.

Your tools. Plain Python splits the document into sentence-sized chunks, then scores each one the same ‘shared words’ way as before.

Investigate. Press Run. The orange bar is the chunk that best answers the question.

Splitting into chunks is how AI can search a huge document without reading the whole thing at once.
Change the query and see a different chunk light up.

your_code.py

Starting Python…

import matplotlib.pyplot as plt

# A longer "document" - too long to search all at once, so we CHUNK it first.
document = (
    "Sharks have skeletons made of cartilage, not bone. "
    "Cartilage is lighter and more flexible than bone. "
    "Sharks can detect a single drop of blood in a swimming pool. "
    "Most sharks must keep swimming to breathe. "
    "A group of sharks is called a shoal or a shiver."
)
chunks = [c.strip().rstrip(".") for c in document.split(". ") if c.strip()]

query = "how do sharks breathe"   # <-- try your own question!
q_words = set(query.lower().split())

scores = []
for chunk in chunks:
    chunk_words = set(chunk.lower().split())
    scores.append(len(q_words & chunk_words))

best_i = scores.index(max(scores))
print("Query:", query)
for i, (c, s) in enumerate(zip(chunks, scores)):
    marker = "  <-- best match" if i == best_i else ""
    print(f"[{s}] {c}{marker}")

colors = ["#ff7a59" if i == best_i else "#2f6df6" for i in range(len(chunks))]
plt.barh([f"chunk {i+1}" for i in range(len(chunks))], scores, color=colors)
plt.title("Ranking each chunk against the question")
plt.xlabel("Words in common")
plt.gca().invert_yaxis()
plt.tight_layout()
plt.show()

The Reveal

Starting Python… (first time only — a few seconds)

🗒️ Detective's notes

Bundles your findings, chart and code into one PDF you can keep, print or hand in.

🧩 How this code works

The tools it uses

Matplotlib — your crayon box — it turns numbers into charts and pictures.

What this one does

• It repeats a step many times with a loop.

How to read it, top to bottom

1. Import — bring in the tools (listed above).
2. The data — the lists and numbers near the top. These are the bits you change.
3. The work — a little maths, or a loop that repeats a step.
4. Show it — print(...) writes words; plt.show() draws the picture.

Use it for your own problems

Change the numbers at the top and press Run — nothing breaks, so experiment!
Ask your own version of the question — swap in your family, your scores, your week.
The same tools work on any numbers — that's the superpower you're learning.

Need a hint?

Each chunk gets its own score, just like the notes did before — we just have more, smaller notes to check.