Homework — Databases

Estimated time: ~50 min · Almost no code · A browser, a free account or two, and a notebook

Three tool-tour exercises. The point is to touch three very different shapes of database in your own browser: a classic SQL engine, a document store, and a vector / similarity-search world. By the end you'll have run an actual SQL query, opened a real JSON document inside MongoDB, and seen — without writing a line of code — what makes vector search different from keyword search.

Before you start

Keep a one-page notebook open (paper or a text file). For each task, write the short answers under (a)/(b)/(c). You are not turning this in — this is the muscle of poking at a database, not just reading about one. The only "code" in this homework is modifying one clause of one pre-written SQL query in Task 1.

The three tasks

Task 1 — Touch a SQL database in the browser

Open sqliteonline.com (or db-fiddle.com if you prefer Postgres/MySQL). Pick the demo Chinook database, or any sample dataset that comes preloaded — most of these sites ship one. Run the example query that's already sitting in the editor; you should see a small table of results appear below it.

Now change one thing. You do not need to know SQL. Examples of a single-word change:

add a WHERE clause, e.g. WHERE Country = 'USA'
change LIMIT 10 to LIMIT 100
change the table name in FROM to a different table from the schema panel on the left

Re-run the query. Notice that the result table changes shape.

Self-check — answer all three:

(a) How many rows did the original query return?
(b) How many rows did your modified query return?
(c) Take a screenshot of one query and its result table for your own notes.

✓ You do NOT need to know SQL — you only need to see that a query produced a table, and that changing one clause changed the table.

Task 2 — Tour MongoDB Atlas (free tier)

Go to cloud.mongodb.com and sign up. The M0 free tier needs no credit card. Create a free cluster — call it landscape — and wait until it shows Active (usually 1–3 minutes).

Then click Browse Collections → Load Sample Dataset. Wait another ~2 minutes for it to import. Once it's done, open the sample_mflix database → movies collection. Click any movie document to expand it — you'll see a tree of fields like title, year, cast, genres. Click a few more documents and notice their fields are not identical — some have awards, some don't; some have tomatoes, some don't.

Self-check — answer all three:

(a) One sentence: how does this differ from a SQL table? (Hint: think about the shape of each row.)
(b) How many documents are in the movies collection? (Atlas shows the count in the collection header.)
(c) Take a screenshot of one document expanded so its JSON shape is visible.

Cleanup

When you're done, you can delete the cluster from Database → ... → Terminate. The M0 free tier is fine to keep running — it never charges — but if you don't expect to come back, tear it down.

Task 3 — See vector / similarity search in action

Open claude.ai (or chat.openai.com or any chatbot you have access to). Ask:

"Find me five animals that are conceptually similar to an octopus but are not cephalopods."

Read the answer. Notice what just happened: the model didn't grep Wikipedia for the literal string "octopus". It found things that are similar in meaning — intelligent, soft-bodied, aquatic, problem-solving. That's the core of how a vector database feels from the outside.

Then open github.com/pgvector/pgvector. Read just the Getting Started section of the README — don't try to install anything, just read. Pay attention to the <-> operator in the example queries.

Self-check — answer all three:

(a) Does the chatbot's answer feel like keyword search or like similarity search? One sentence on why.
(b) What does the <-> operator do in pgvector? (One short sentence from the README.)
(c) Write one sentence on why a vector database is a natural fit for retrieval-augmented generation (RAG). Lesson 13 returns to this topic in depth.

What "done" looks like

You're done when, in your notebook, you have:

three row-count numbers and one screenshot from Task 1 (SQL)
one shape-difference sentence, one document count, and one screenshot from Task 2 (MongoDB)
three short sentences from Task 3 (vector / similarity)

The aim was never to make you a DBA in 50 minutes. It was to make sure that the next time someone says "we'll just shove it in Postgres" or "that needs a vector store", you have a picture in your head of what they're actually pointing at — because you've now opened all three.

Where this goes next

Module 3 puts data in motion: how it travels over the network (Lesson 10), how systems talk to each other (Lesson 11), and how all of this lives in the cloud (Lesson 12). Module 4 then comes back to databases through the AI lens — embeddings, vector indexes, and RAG — so Task 3 is a foretaste, not a deep dive.