Just ask: designing intent-driven algos
summary
Claims that capturing real-time intent closes the gap between a user’s general profile and what they want right now and demonstrates this by rebuilding internal recommenders. Uses explicit free‑text intent, a style shuffle thumbs system, batch predictions plus a live constrained optimizer implemented with Pyomo to choose inventory counts. Built a buyer-facing tool that returns feasible assortments, speeds up iteration, and improves decisions at scale while introducing latency and UX challenges; over 100 people work on related algorithms.
AI-written summary · about
chapters (10)
- 0:00Intent-driven recommendation systems
- 2:16Stitch Fix: human + algorithms
- 6:06Limitations of filtering methods
- 8:17Asking for user preferences
- 10:22Recommenders for internal buyers
- 12:30From filtering to optimization
- 15:28Intent to optimized recs
- 17:37Mapping and routing parallels
- 19:53Flexible recommendation scenarios
- 21:03Q&A
related talks
28mThe Power of Personalized Customer Experiences Through AI Innovations
41mRecommender Systems: Crafting Effective Personalised Experiences
27mTransform online shopping experiences with personalized recommendations
18mSynthesizing human and machine capabilities
38mOptimizing Retail Forecasting Through AI Models
42mAn Algorithm of Style
38mRecommendations at Rappi: from MVP to Data Product
27mEnhancing Personalization Using Recommendation Systems & Multi-Armed Bandits