Comet

Comet provides tools that assist in the development of machine learning (ML) and deep learning (DL) models for other organizations and enterprises. Comet was founded in 2017 by Gideon Mendels and Nimrod Lahav. The company is headquartered in New York City with offices in Tel Aviv, Israel.

Comet is able to be used for both standalone experiment tracking and model production monitoring. ML models can be tracked, compared, debugged, and monitored from their earliest experimentation phase all the way through the production stage. Comet Model Production Monitoring (MPM) tracks data drift and accuracy metrics of models in production. The model registry allows users to easily track changes throughout a model's lifecycle and compare the performances of various models in production. Comparisons can be made between model codes, hyperparameters, metrics, predictions, dependencies, and system metrics. Model datasets can be saved on Comet Artifacts for easy visibility and collaboration. They can be linked to models in training and production. Comet can be run on any infrastructure including on-premises, cloud, and virtual private cloud (VPC).

In November 2022, Comet announced a new product called Kangas, an open-source tool for real-time, large-scale data visualization.

Comet serves hundreds of companies, including Shopify, Affirm, Etsy, Ancestry, Uber, Zappos, Cepsa, and AssemblyAI. An estimated 70 percent of the North American population has interacted with a machine or deep learning model trained by Comet.