Cheaper, faster machine learning

Meeshkan is a framework for accelerated machine learning. It uses JIT compilation, a real-time scheduler, and state-of-the-art gradient compression to squeeze as much juice from your processing units as possible.

Slash your ML budget

Some teams have unlimited resources and time to do machine learning. For those that do not, there's Meeshkan.


Run several notebooks or jobs on one GPU using Meeshkan's best-in-show scheduling. Effortlessly scale up to multiple GPUs

On premise

Cloud GPUs cost a lot of money. Buy your own hardware and deploy Meeshkan on-premise for deep savings.


Minimally invasive. Activate Meeshkan with one line of code in your Jupyter or Zeplin notebooks and your batch scripts.

How it works


Step one

Write machine learning jobs in your favorite environment. Meeshkan takes care of scheduling out-of-the-box without any additional code.


Step two

Change jobs' priority in real-time and keep track of algorithms' loss and accuracy with alerts. If you are linked to a cloud service, Meeshkan will automatically scale up or down servers based on your preferences.


Step three

Voyage through time with our intuitive analytics tools to understand what you've accomplished and where you're going. Easily build upon previous work with transfer learning.

Get Started