This is a great example of where LLMs complement vector search rather than replace it. Embeddings are excellent for semantic matching, but eCommerce queries often contain structured intent like brand, size, color, model, or quantity that should become filters. Using an LLM to extract those attributes before running a hybrid search can significantly improve relevance. The main challenge is balancing accuracy, latency, and cost, especially at production scale.
This is a great example of where LLMs complement vector search rather than replace it. Embeddings are excellent for semantic matching, but eCommerce queries often contain structured intent like brand, size, color, model, or quantity that should become filters. Using an LLM to extract those attributes before running a hybrid search can significantly improve relevance. The main challenge is balancing accuracy, latency, and cost, especially at production scale.