Lore is a machine learning framework from Instacart that aims to make ML approachable for software engineers and maintainable for ML researchers. It provides a consistent project structure and abstractions for data, features, models and predictions, so a model can move from a notebook experiment to a production service without being rewritten. It is an open-source project on GitHub with around 1,500 stars, developed during Instacart’s ML platform growth.
Many ML codebases fail at the handoff between research and engineering: training code, feature logic and serving code drift apart. Lore addresses that with conventions and reusable components – datasets and encoders that behave identically in training and prediction, model wrappers around popular libraries, and built-in support for experiment tracking, validation and deployment patterns. The framework reflects lessons from running recommendation and forecasting models at grocery-delivery scale, packaged so smaller teams can adopt the same discipline.
Lore is free and open source under the MIT license.
Software engineers and ML teams who want a disciplined, production-minded framework for building and maintaining Python ML systems.
Lore was open-sourced by Instacart in 2018 and reflects the company’s experience scaling machine learning for search ranking, recommendations and logistics. Its opinionated structure – datasets, encoders, models and predictions as named, reusable pieces – was an early answer to a problem the industry later attacked with feature stores and ML platforms, and the codebase remains a readable reference for teams designing their own internal ML conventions.