AI Search for Marketplaces: Experiences & Tech
summary
Velebit AI presents deployed features: search by image using a custom CNN with binary vectors in production since 2016 for fashion, a recent multimodal vectors pipeline (CLIP-style) to match text and images, and a LLM chat-based query refinement that parses conversation, extracts brand/size, and iteratively updates vector queries. Also built mobile realtime object detection for category and price hints, attribute tagging APIs, duplicate detection via image embeddings, and hybrid retrieval with reranking (cross-encoders) and vector DBs for 2,000–3,000 category taxonomies.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:09AI search solutions overview
- 4:20E-commerce challenges & trends
- 5:25Buyer-focused search use cases
- 8:15Fashion search with multimodal AI
- 10:01Chat-based search with LLMs
- 11:22Real-time object detection
- 12:26Seller product enhancements
- 16:00AI search technical trends
- 20:42Building multimodal search systems
- 22:29Attribute extraction techniques
- 26:44Semantic intent & hybrid search
- 28:51Hybrid fusion of search results
- 30:15Conclusion & personalized search
related talks
30mEvolving Search in Online Marketplaces: Practical Applications of AI
29mDeveloping Visual AI Solutions for Online Marketplace
16mSolve for AI
33mMachine Learning in Visual Search
41mAI is here – time to throw away our search engines?
40mAI or the Metaverse - What should brands bet on?
41mSearching for Products by Photo
33m"Are You Sure?": blending product comparisons and recommendations with A.I.