Layer: Declarative MLOps Platform for ML Applications at Scale
Thrilled to invest in Layer and be a part of their story together with Hummingbird Ventures and FJ Labs!?
Lack of readily available infrastructure to operationalize and collaborate on models
Number of ML practitioners is growing exponentially, but there still is no collaborative platform that makes it easier to collaborate on models. Evaluation and feedback loops are too long with the existing ML workflows. And by the time ML applications are released, initial requirements and designs become obsolete.?
These problems can be solved through a real-time collaboration platform with a diverse feature set encompassing MLOps value chain, including data and model catalogues, versioning, monitoring to manage degradation by potential data drifts, and reusable libraries.?
Here comes Layer: Github for ML practitioners
Layer is building a Declarative MLOps platform to help data teams produce ML applications in small and safe increments that can be reproduced, tested, and reliably released at any time, in short adaptation cycles. MLOps is instrumental in bringing ML system workflows to production. It leads to less friction among teams and corresponding operations teams.
Layer orchestrates a practitioner’s data and model pipeline for her and empowers her to focus on building models without worrying about the infrastructure. It supports the majority of ML frameworks including TensorFlow, SparkML, PyTorch, and scikit-learn.
Central platform to handle it all
Layer integrates with Github to automatically build the entities (features, model, etc) from code revisions committed to Github while the model is run on its own platform and monitored through a customizable dashboard. It enables collaboration on both models and data inputs including the usage of data catalogs (data discovery, feature store, automated building, and testing), model catalogs (model registry, testing, performance monitoring, deployment, hyperparameter tuning), and data lineage.
领英推荐
Talent is scarce and the state-of-the-art is evolving rapidly
ML adoption grows exponentially due to its obvious benefits, but the limited number of ML practitioners, together with the lack of infrastructure readily available to operationalize models are the biggest bottlenecks. For many enterprises, running machine learning in production has been out of the realm of possibility.?
Layer is built to tackle the efficiency problem in the data science space and to democratize access to ML applications by empowering data teams. It competes with a number of players including Tecton, Google Vertex AI, Amazon Sagemaker, Datadog, and Databricks. Its competitive differentiation includes:
Layer has a bottom-up, open-source, and community-first approach targeting to create a collaborative platform and leverage network effects. The collaborators act as “same-side” network bridges, organically attracting new users to the free plan, as well as converting them to paid plans when the intra-company network expands beyond freemium.
Mehmet has built an all-star team?
The bar for building such a platform is high, but Layer’s team got it all covered. Mehmet is a successful repeat founder, having previously built, scaled & exited Gram Games to Zynga. He has also brought together a stellar team, with senior ML executives and builders from Palantir, Coinbase, Revolut, Datadog, Hazelcast, and the likes.
Layer has already built a top-quality infrastructure. The company aims to combine SaaS and network effects in building up a community of practitioners and monetize with an enterprise-ready end-to-end product. We are proud to join their most recent round and are truly thrilled to partner with Mehmet and the rest of the team on this journey.
Enis, gracias por compartir!