Vector Embeddings Made Simple: A Kitchen Shelf Analogy

Scene:
It’s a lazy Sunday afternoon. I’m sitting in the kitchen, sipping tea, when Ma walks in, arranging groceries.

Mom:
“Shuvo, why are all these packets mixed up? Sugar, salt, pasta — shouldn’t they be in separate places?”

Me:
“Separate places, why?”

Mom:
“Yes, because it will be easier to find the right”

Me:
“Wait, Mom, I’m about to tell you something straight from the world of AI. Trust me, it’s well-organized information.”

Mom:
“AI? You mean robots?”

Me:
“Sort of… but not the kind that walk around. This is more about how computers think.”

Mom:
“Do computers think? Since when? I thought they just follow instructions.”

Me:
“They do, but sometimes we want them to do more than just follow orders. We want them to figure out relationships between things. Like how you know sugar and salt go in the same cupboard, and coffee and tea belong together.”

Mom:
“Hmm… so you’re saying this is about organizing? I can understand that. I’ve been organizing this kitchen for years.”

Me:
“Exactly! And that’s why I thought of you. The way you arrange groceries is basically what computers do when they create something called a ‘meaning map’.”

Mom:
“A meaning map? Sounds like something Google Maps would make for the kitchen.”

Me:
“Haha, not quite — but you’re not far off…”

Me:
“Okay, imagine a giant invisible space in your mind. Every item you know — sugar, salt, coffee, cat, dog — has a position there.

• If two things are similar, they’re close together.

• If they’re different, they’re far apart.

When you think of ‘cat’, you instantly think of ‘kitten’, ‘meow’, and ‘pet’. That’s because in your brain, those things are stored close together. Computers can’t do that naturally — so we teach them using something called vector embeddings.”

Mom:
“Alright… but how does a computer understand this?”

Me:
“We turn every idea into a list of numbers — kind of like giving it an address on an invisible map.

For example:

• Coffee’s address might be (2, 3)

• Tea’s address might be (2, 4)

• Banana’s address might be (10, 1)

Since coffee and tea are close on the map → they’re similar, both drinks.
Bananas are far away → different, it’s a fruit.”

Mom:
“So you’re saying that when I arrange groceries, I’m working like a computer?”

Me:
“Exactly! You’re doing vector embeddings with your pantry. The only difference is, computers use math and lists of numbers, while you use common sense.”

Why Computers Need This

Me:
“Computers don’t ‘understand’ coffee or cats the way we do. To them, it’s all just data. But with embeddings, they can:

• Find similar things

• Group related things together

• Search smartly instead of just matching exact words

That’s why your phone can find all photos of our cat even if you never tagged them ‘cat’. It’s walking through its meaning map and saying, ‘Hmm… this looks close to cat, so I’ll include it.’”

Mom:
“So basically, if I can organize tea, coffee, and sugar, I could teach a computer, too?”

Me:
“Mom. If computers had mothers, they’d probably learn embeddings from them.”