Back to work

Case Study

Semantic Retrieval Platform

A search layer combining embeddings, query rewriting, and taxonomy constraints for high-signal discovery.

Challenge

Keyword-only lookup struggled with domain vocabulary and intent mismatch across internal knowledge bases.

Approach

Used pgVector with OpenAI embeddings plus taxonomy-aware post filters for accurate top-K retrieval.

Period

2024 - 2025

PostgreSQLpgVectorOpenAIJavaSystem Design

Outcomes

  • Improved knowledge discovery efficiency by 40%.
  • Made search quality more stable for ambiguous and long-form queries.
  • Created a strong foundation for AI-assisted support workflows.