Optimizing Search Advertising with Semantic Search Using Vector DB

A leading digital marketing company that connects advertisers with audiences through a marketplace​

Challenges​

  • Traditional text searches lacked contextual understanding, relying only on keyword-based approaches​
  • High latency in ad delivery led to missed opportunities.​
  • Required uncompromised performance for large-scale datasets (search queries, user logs) in real-time.​

Solutions

  • Data Clustering: Multi-model approach with product-category-specific models for 65M products across 30K categories.​
  • Vector DB & Indexing: High-dimensional vectorization for efficient semantic search.​
  • AI-Powered Selection: Automatic SLM selection for contextual correctness using semantic similarity and relevance scoring.​
  • GPU-Based Optimization: Improved query vectorization for performance at scale.

Impact

  • earch query time reduced to <100ms (including network time).​
  • 30% increase in CTR due to enhanced ad relevancy.​
  • Cost benefits: ZERO GPT subscription costs (via standalone finetuned SLMs) and efficient indexing reducing compute overhead.​
  • Improved search performance with low latency.​

Related Stories

Unlock faster growth & intelligence.