a hands-on course

Build a Semantic
Search Engine

(Embeddings & Semantic Search)

Search text by meaning, not keywords — build a working semantic search engine in about half an hour.

A free, hands-on tutorial: build a semantic search engine that finds text by meaning instead of matching keywords. Turn sentences into embeddings with a local model, rank them by cosine similarity, and search a corpus — the retrieval engine behind modern search and RAG. Runs entirely on your laptop, no API key.

in the real world

Semantic search powers help centers, site and code search, and the retrieval step in RAG chatbots — finding the right passage even when it shares no words with your query.

Start with the setup →

Part 1 — Set up your machine

1

Install the model

A clean virtual environment, one pip install, and a script that loads a real embedding model.

2

Turn text into vectors

Encode a sentence into an embedding — a list of numbers that captures its meaning — and see similar sentences land close.

Part 2 — Build the search engine

3

Embed your corpus

Collect the documents you want to search and encode them all into one matrix of embeddings.

4

Rank by cosine similarity

Measure how close two embeddings point — the cosine of the angle between them — to find the best match.

5

Build the search function

Wrap it into a search(query, k) that returns the top matches — and watch it find meaning, not keywords.

Part 3 — Search it and scale up

6

Persist, scale & what's next

Save embeddings to disk, learn when you need a vector database — and how this becomes the retrieval in RAG.