AI-Powered Search in Modern E-Commerce
summary
A custom two-tower model trained on millions of queries and historical clicks uses contrastive loss to create domain embeddings and delivers a 21% increase in CTR over keyword and generic embeddings; the pipeline stores product vectors in a vector database, uses bi-encoders for fast candidate retrieval and cross-encoders or a learned reranker for slow exact ranking, leverages LLMs to synthesize cold-start training data, and balances batch embedding updates with real-time availability while monitoring query types and latency trade-offs.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 2:34AI search promise in retail
- 5:47Traditional keyword limits
- 8:15Multimodal embeddings
- 11:46Query routing algorithms
- 14:38Personalization next layer
- 18:11Fine-tuning embeddings
- 21:47Conversational shopping
- 25:11Q&A
related talks
33mAI-Powered Search in Modern E-Commerce
22mMachine Learning: The Engine Behind Modern E-commerce Search
31mAdvanced search for your e-commerce website
13mThe State of AI Powered Search and Retrieval
22mAI-Powered Personalized Discovery with Graph Search and VLM-Enhanced Embeddings
27mMachine Learning for Advanced AI Search
22mTensor arithmetics in search and ranking for Ecommerce.
44mRobust AI Search Ranking for Radical C2C Marketplace Growth