Vector search for web developers: clear, useful, and kind of mind-blowing

For the last 20 years, we've built websites that search for words.

SELECT * FROM articles WHERE content ILIKE '%massage%'

Now, we can build websites that search for meaning.

And when I discovered how simple it was to implement --- and how powerful it felt in use --- I realized this could redefine how we write code, but also how we shape user experience.

---

Quick backstory

Vector-based search isn't that new.

The first ideas of "vector space models" in information retrieval date back to the 1970s. But the real breakthrough came when we started to represent text by its meaning, not its exact words.

Not with keywords. With vectors.

That became possible with the rise of powerful AI models like BERT (Google, 2018), GPT (OpenAI, 2019+), and more recently, general-purpose embedding models like text-embedding-3-large.

---

What is vector search, really?

It's a way to retrieve content that's semantically close to a query, not just textually close.

A question like "Should I drink water after the treatment?" might pull up "I recommend staying well hydrated after a massage."

No overlapping words --- but the meaning is spot on. And the machine gets it.

---

How it works (developer version)

1. Each piece of text is turned into a numeric vector (embedding) → 3072 values if you're using text-embedding-3-large

2. The vector is stored in your database (e.g., PostgreSQL with the pgvector extension)

3. When a user asks a question, that query is also embedded → And you search for the closest vectors (cosine similarity, Euclidean distance...)

4. You return the most semantically relevant content → Optionally, GPT can rephrase it nicely.

All this can happen in milliseconds.

---

Where you're already using it (maybe)

You're likely already interacting with vector search:

• YouTube → video suggestions
• Spotify → music recommendations
• ChatGPT / Notion AI / Perplexity → hybrid retrieval
• Amazon / Shopify → similar product matching
• Modern chatbots → contextual FAQ and support

---

Use cases for business websites

Here are a few concrete examples:

• A service business builds a smart FAQ that understands question variations
• An e-commerce store can respond to: "Is this sweater warm enough for winter?"
• An agency integrates an assistant that understands technical questions, even poorly phrased ones
• A content platform can retrieve relevant articles, even if the user misses the right keywords

All of this, without ElasticSearch or custom NLP pipelines.

Just:

• OpenAI for the embeddings
• pgvector for storage
• A well-crafted SQL query

---

Why I added it to my projects

I build AI assistants --- for businesses, and sometimes for people who want to leave a legacy.

In those projects, users aren't looking for words. They're looking for answers, for meaning, for a real interaction.

And the day I connected my vector database and saw the assistant respond meaningfully to a never-before-seen question --- I knew I wasn't going back.

---

What it really changes

• You're no longer coding word filters --- you're building understanding
• You're not sorting by tags --- you're sorting by intent
• You're not just designing visual UX --- you're building intelligent UX

And it doesn't take six months of R&D. It takes the willingness to add a semantic layer to what you create.

---

For those who want to try it out

If you can write a SQL query, you can implement vector search this week.

Here are a few key tools:

If you're curious how I integrated it into my assistants --- feel free to reach out.

---