Originally published on Medium.
Ever wonder how YouTube just knows what you’ll want to watch next? Or how Instagram recommends just the reel that keeps you scrolling past midnight?
It’s not coincidence. It’s embeddings — the hidden language behind how systems “understand us.”
How Embeddings Are Created
At a high level, embeddings are vectors — lists of numbers — that capture relationships and meaning.
When a model analyzes text, images, or products, it converts them into a point in a multidimensional space.
Items that are “meaningfully similar” end up close together — e.g. “coffee mug” sits near “tea cup,” but far from “motorcycle.”
Bringing That Power Into My Project
In my current e-commerce build, I use MongoDB + Voyage AI to generate embeddings for every product.
Instead of just indexing item names or tags, I store dense vectors — so when someone searches “affordable office laptop,”
the system doesn’t merely match keywords. It surfaces products that feel relevant.
Those long lists of numbers? They might look odd, but to the model, they encode relationships, context, and meaning.
Every smart system — whether it’s YouTube, Instagram, Spotify or Reddit — is built on the same foundation:
embeddings, silent but powerful.
For me, implementing even a small slice of that in my project transformed how I think about code, systems, and intelligence.
Embeddings might not trend on social, but they are the quiet revolution behind everything that feels smart online.
And being on the journey to “speak” that language? That’s what makes building things feel alive.
Stay tuned — I’ll drop a link to my project soon.
Top comments (0)