· RAG · Embeddings · PostgreSQL
Explaining embeddings with a 3D map (because I had to understand them too)
Two repos I built to learn RAG in public. A visual embeddings explorer and an end-to-end tutorial on the Pagila database.
When you build RAG at work you learn fast, but in bits and pieces. So between late July and early August 2026 I made two repos to organize what I’d learned and, along the way, explain it to someone else.
Que-son-los-Embeddings (“What are embeddings”)
On July 28 I pushed Que-son-los-Embeddings-. It’s a Flask app with Qdrant where you type sentences and see them on a 2D or 3D map (with Plotly). Sentences that mean similar things end up close together, even if they don’t share a single word.
That’s the whole idea of embeddings: turning text into a vector of numbers where “close” means “similar”. Watching it spin in 3D makes it click in two seconds, which the textbook definition never did for me.
pagila-rag-tutorial
On August 5 I started pagila-rag-tutorial: an end-to-end RAG pipeline over Pagila, the sample movie-rental database for PostgreSQL.
The funny part: the first commit says “Corrected Pagila schema, data and commands”. Before doing any AI I had to fix the sample database because it didn’t load properly. Welcome to real life.
The next day I reorganized it into numbered folders (step 1, step 2…) so anyone can follow it in order: load the data, generate embeddings, store them in PostgreSQL and run semantic searches like “a horror movie with a dog”.
Why build “teaching” repos
Because explaining something forces you to really understand it. Several things I do today in Xerebro only fully clicked while building these tutorials, not the other way around.