10 posts · searchable
10 posts
Describes how Lyft improved the pickup experience for riders in gated communities by addressing communication and navigation challenges.
Explains the use of Bayesian trees to predict rider conversion in sparse data environments at Lyft.
Discusses using surrogacy and region-splits as alternatives to A/B testing for measuring long-term marketplace effects.
Details how Lyft scaled its localization efforts by integrating AI to replace or augment human translation.
Covers validation and diagnostic techniques for doubly robust models used in causal inference beyond randomized experiments.
Describes the architecture, optimization, and evolution of Lyft's Feature Store for machine learning.
Chronicles the journey of upgrading from Python 3.8 to 3.10 and debugging a memory leak encountered along the way.
Explains the evolution of Lyft's ML platform architecture, LyftLearn, to support thousands of production models.
A personal account of a starter project on the Lyft Rider Data Science team using the Rider Experience Score tool.
Describes the multi-year migration of Lyft's Android codebase from Java to Kotlin across multiple apps.