Recommendations at Rappi: from MVP to Data Product
summary
Describes a working pipeline that moves from simple baselines to a production recommender: start with random and top-sold lists, establish a baseline, then add statistical and ML models and finally personalization. Claims an end-to-end stack with Airflow for ETLs and retraining, embedding-based item and user representations (fastText/word2vec style), a real-time feature store for online predictions, and standard A/B testing to tie offline metrics to business lift. Emphasizes UX changes and incremental experiments as the levers that actually move conversion.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:32Machine learning in production
- 3:18Defining a data product
- 5:50Data product misconceptions
- 7:58Data quality strategy
- 10:12Building with minimum viable products
- 12:44Baselines and research timeboxing
- 14:35Experiment design validation
- 16:26Business vs model metrics
- 18:35Complex models feasibility
- 20:21Recommendation system requirements
- 23:15Recommendation engine foundations
- 26:31Personalization challenges
- 28:26Rappi tools infrastructure
- 30:33Evolution of recommendations
- 33:05Q&A
related talks
28mThe Power of Personalized Customer Experiences Through AI Innovations
41mRecommender Systems: Crafting Effective Personalised Experiences
58mRecommender Systems – A Walkthrough
23mFrom Data to Recommendations: Building an Intelligent System with ROR
27mEnhancing Personalization Using Recommendation Systems & Multi-Armed Bandits
32mResurrecting a Recommendations Platform
43mRecommendation Systems in Market Research
29mThe Challenges of Building a Personalized Recommendation System in E-Commerce