All labs · Lab 20 · beginner · 30 min

Search by meaning

Turn text into vectors and find the closest sentences, first in memory and then in a database.

You build: search.py: finds the right sentence even when it shares no words with the question.

  • Python 3.11 or newer and uv
  • Ollama
  • Docker
Ollama docspgvector

  1. 01Pull an embedding model and embed one sentence

    ollama pull nomic-embed-text && curl -s http://localhost:11434/api/embed -d '{"model":"nomic-embed-text","input":"my box arrived crushed"}' | head -c 200

    You pass when: JSON with an embeddings array of numbers.

  2. 02Write 30 sentences about parcels and rank them against a question by cosine similarity

    uv run --with numpy --with requests search.py "my box arrived crushed"

    You pass when: The top three are about damage, though none of them contains the word crushed.

  3. 03Start Postgres with pgvector

    docker run -d --name learn-pg -e POSTGRES_PASSWORD=learn -p 5433:5432 pgvector/pgvector:pg17 && sleep 5 && docker exec learn-pg psql -U postgres -c "CREATE EXTENSION IF NOT EXISTS vector; SELECT extversion FROM pg_extension WHERE extname='vector';"

    You pass when: One row with the extension version.

  4. 04Store the vectors in a table and run the same search in SQL

    uv run --with numpy --with requests --with "psycopg[binary]" search.py --db "my box arrived crushed"

    You pass when: The same top three as the in-memory search.

The check runs on the computer, not on your word: a step passes when its command exits with 0 there.

Your computer · Fleet · off

Computers are not set up on this site yet. Run each command on your own machine and compare with "You pass when".