One search. Connected ideas.

A vector database stores vectors and supports similarity queries with metadata. Vector search is the numerical operation; semantic search is the application goal of finding meaningfully related content. Existing databases can also add vector indexes, so you…

Beginner — Why use a vector index?

Pinecone, Qdrant, Weaviate, pgvector, MongoDB Vector Search, Elasticsearch, and OpenSearch provide different ways to store and retrieve vector-backed knowledge. Some are dedicated vector services; others extend databases or search engines you may already…

Atlas is MongoDB's managed database environment; MongoDB Vector Search indexes supported vector fields for similarity retrieval. Source metadata and application records can live alongside chunk documents. A vector/search index is distinct from ordinary…

Queues decouple arrival from processing. Caches reduce repeated eligible work. Business databases store authoritative records; vector indexes store derived retrieval representations. Each has a distinct consistency and recovery contract.

AI and LLM fundamentals

Keyword search matches words or indexed terms rather than learned meaning. BM25 is a common relevance formula that rewards useful matching terms, accounts for document length, and avoids letting repeated words dominate indefinitely. It is usually implemented…

Hybrid search combines lexical and semantic candidate retrieval. Metadata filtering narrows records by explicit attributes such as tenant, policy version, date, or region. Filtering answers “which records are eligible,” while ranking answers “which eligible…

Reranking is a second ranking stage that examines a short candidate list more carefully. A reranker often reads the query and each candidate together, allowing richer comparison than separate embeddings. It changes candidate order; it cannot recover a…

Requirements: Upload approved PDFs, track ingestion, ask questions, return source references, and abstain when evidence is missing.

Use this after studying the full explanations. Spend 10 minutes on vocabulary, 10 on RAG and agents, 10 on security/workflows/evaluation, and 10–30 on the insurance design and scenario answers.

Similarity measures how close two vectors are according to a selected metric. Cosine similarity compares their direction, effectively asking whether they point in a similar direction regardless of magnitude. Dot product and Euclidean distance are other common…

A vector database stores vectors and supports similarity queries with metadata. Vector search is the numerical operation; semantic search is the application goal of finding meaningfully related content. Existing databases can also add vector indexes, so you…

I first test the existing database's vector capability against filtered recall, latency, lifecycle, security, and cost needs. A dedicated store is justified when it solves measured limitations enough to offset another operational dependency.

Pinecone, Qdrant, Weaviate, pgvector, MongoDB Vector Search, Elasticsearch, and OpenSearch provide different ways to store and retrieve vector-backed knowledge. Some are dedicated vector services; others extend databases or search engines you may already…

Beginner — Why use a vector index?

Exact search examines eligible vectors thoroughly; approximate search trades some nearest-neighbor recall for speed and scale. I benchmark with realistic filters, corpus sizes, and query patterns rather than assuming ANN is always required.

Atlas is MongoDB's managed database environment; MongoDB Vector Search indexes supported vector fields for similarity retrieval. Source metadata and application records can live alongside chunk documents. A vector/search index is distinct from ordinary…

It makes repeated similarity queries efficient over a growing corpus. Small datasets can use exact comparison, but larger collections need indexing and careful metadata filtering for relevance and eligibility.

The retrieval process selected candidates from an ineligible corpus. I move supported eligibility filters into retrieval and recheck object permissions before evidence reaches the model. Post-filtering alone can both underfill results and expose data upstream.

Keyword search matches words or indexed terms rather than learned meaning. BM25 is a common relevance formula that rewards useful matching terms, accounts for document length, and avoids letting repeated words dominate indefinitely. It is usually implemented…

Hybrid search combines lexical and semantic candidate retrieval. Metadata filtering narrows records by explicit attributes such as tenant, policy version, date, or region. Filtering answers “which records are eligible,” while ranking answers “which eligible…

Reranking is a second ranking stage that examines a short candidate list more carefully. A reranker often reads the query and each candidate together, allowing richer comparison than separate embeddings. It changes candidate order; it cannot recover a…

Queues decouple arrival from processing. Caches reduce repeated eligible work. Business databases store authoritative records; vector indexes store derived retrieval representations. Each has a distinct consistency and recovery contract.

AI Engineering learning module

AI Engineering learning module

AI Engineering learning module

AI Engineering learning module

AI Engineering learning module

AI Engineering learning module

AI Engineering learning module

AI Engineering learning module