Vector Databases & ANN Search — Core

What a vector DB is, ANN vs exact search, HNSW/IVF/PQ indexes, distance metrics, metadata filtering, quantization, and vendor concepts (Pinecone, Qdrant, Milvus, Weaviate, Chroma).

Medium50 questions

Sample questions

What is the core job of a vector database?

To store embeddings and find nearest vectors by similarity
To train the embedding model that produces the vectors
To translate SQL queries into natural language
To compress images into thumbnails for display

Why

A vector database indexes embedding vectors and serves fast nearest-neighbor (similarity) search, usually with metadata filtering and CRUD. It doesn't train the embedding model, translate SQL, or process images.

Pinecone: What is a vector database Qdrant: What is a vector database

How does a vector database differ from a standalone ANN library like FAISS?

It wraps ANN with persistence, filtering, scaling
It cannot perform nearest-neighbor search at all
It only stores text, never vectors
It replaces the need for embeddings entirely

Why

A vector DB wraps ANN search with production features: durable storage, payload/metadata filtering, distributed sharding/replication, security, and live updates — beyond a raw index library. It does do ANN search, stores vectors, and still needs embeddings.

Qdrant: What is a vector database FAISS: Billion-scale similarity search

Why do vector databases use Approximate Nearest Neighbor (ANN) search instead of exact search?

Exact search over millions is too slow
Exact search changes the stored vectors
ANN returns more results than exact search
ANN is required to compute cosine similarity

Why

Exact nearest-neighbor compares the query against every vector (linear cost), which is too slow at scale; ANN trades a little recall for sublinear search. It doesn't alter vectors, return more results, or enable cosine.

FAISS: Billion-scale similarity search Pinecone: What is a vector database

HNSW, a popular ANN index, is best described as what kind of structure?

A single flat list scanned linearly
A hash table mapping vectors to exact buckets
A balanced B-tree over scalar keys
A multi-layer graph searched greedily from the top

Why

HNSW builds a hierarchy of proximity graphs: search starts in a sparse top layer and descends, greedily hopping to closer neighbors — fast, high-recall ANN. It is not a flat scan, a hash table (that's LSH), or a B-tree.

HNSW paper Qdrant: What is a vector database

In HNSW, what do the upper (sparser) graph layers do during a search?

Long hops reach the query's region fast
They store the raw document text for each vector
They hold the metadata filters for the query
They compress the vectors to save memory

Why

Sparse upper layers have long-range links, so the search covers big distances fast to reach the query's neighborhood, then refines in denser lower layers. They don't store text, filters, or perform compression.

HNSW paper Pinecone: What is a vector database

Free account

Take the full module

These are the first few of 50 questions. A free account opens the rest as a scored drill.

Every question in this module
Instant feedback and supporting reading
Your score and progress, saved

Free · your email is used for progress only.

About Nybble^™

The AI space moves fast.
Nybble^™ is how you keep up — and stay sharp.

Snack

What happened. In two minutes.

The AI news cycle moves at a pace no one can keep up with. Snack distills what launched, what shipped, and what matters — every day, without the filler.

Go to Snack

Stack

The concepts behind the headlines.

News tells you what. Stack tells you why and how. From RAG architectures to agentic evals, these are the ideas that will shape what you build next.

Go to Stack

Hack

Prove you actually get it.

Reading about LangChain is not the same as knowing it. Hack challenges you with production-grade questions, then shows you the references that make the answer stick.

Go to Hack