Vector DB vs SQL

Vector databases power every RAG system, yet most candidates can't explain ANN algorithms or hybrid search. This fundamental question appears in 80% of AI engineering loops. Master dense vs sparse retrieval and when hybrid search wins.

500M vectors at sub-100ms p99 is a staff-level vector search design question from Uber, Airbnb, and large-scale ML platform teams. Learn sharding strategies, index tuning, and the operational trade-offs that separate senior from principal engineers.

Medium vector DB / embeddings trade-off question on dimensions vs cost/quality.

Hard vector DB question on tenant isolation, noisy neighbors, and ACL safety.

Latency, availability, freshness, recall, and error budget definitions with realistic dependencies.

Latency, recall proxies, ingest lag, index size, filter rates, error budgets, and RAG downstream signals.

Field selection, normalization, ACL tags, temporal fields, and schema evolution without breaking filters.

Brute-force exact search thresholds, ANN recall trade-offs, and hybrid exact re-rank on shortlists.

Company-based system design: sharded ANN, metadata pre-filtering, query routing, reranking, and latency SLOs at marketplace scale.

Encryption at rest/transit, tenant isolation, ACL on metadata filters, PII in embeddings, and audit logging.

Write buffering, mutable segments, async merge, backpressure, and consistency windows for hot ingestion paths.

Learned sparse representations, inverted index integration, lexical expansion, and fusion with dense ANN.

efSearch, nprobe, over-fetch for filters, offline benchmarks, and continuous monitoring of retrieval quality.

Pricing dimensions: vectors stored, dimensions, QPS, namespaces, egress, and when self-host breaks even.

Shared embedding spaces, separate indexes, CLIP-style models, metadata routing, and fusion strategies.

Dual-write, shadow traffic, embedding compatibility, cutover rollback, and validation gates.

PostgreSQL + pgvector tuning: HNSW vs IVFFlat, connection pooling, vacuum, partitioning, and when to outgrow Postgres.

BM25 + vector fusion, RRF, weighted scores, rerankers, and same-chunk alignment pitfalls.

Snapshot strategies, embedding model lineage, cross-region restore, and rebuild-from-source playbooks.

Bootstrap strategies, fallback retrieval, hybrid search defaults, and UX for new tenants or collections.

PCA, Matryoshka embeddings, learned compression, and recall impact when shrinking vector size.

Read replicas, eventual consistency after upserts, quorum writes, and RAG staleness expectations.

Horizontal scaling patterns: hash sharding, semantic partitions, routing embeddings, and merge/rerank at query time.

Hard vector DB topic: filter-aware ANN, post-filter recall collapse, and hybrid query planning for tenant isolation.