Write-up
Designing Semantic Search with pgVector and OpenAI Embeddings
Production notes on top-K retrieval, query rewriting, and taxonomy-aware filtering for better relevance.
Vector SearchOpenAIPostgreSQLSystem Design
Keeping vectors close to transactional data in PostgreSQL simplified operations and reduced synchronization drift between systems.
Query rewriting improved retrieval quality by normalizing intent before similarity search, especially on long enterprise-style queries.
Taxonomy-aware filtering created a better precision-recall balance by constraining similarity checks to context-relevant branches.
This section is designed to scale into longer essays later. Add richer Markdown-style content directly in
src/content/portfolio.tsfor each entry.