Connecting local Jupyter notebooks to remote cloud compute.
Moonglow connects local Jupyter notebooks to remote cloud compute. Machine learning researchers and data scientists use us to scale up their experiments without having to do DevOps. Previously, Leila was a software engineer at Jane Street. She led the build-out of its equities clearinghouse connectivity infrastructure and was the technical lead of its front-office SRE team. Trevor was part of Hazy Research Lab at Stanford, and published machine learning research at NeurIPS, ICLR and ACL. With the explosive growth of deep learning, there are over 1 million ML researchers who do computationally intensive experiments every day. They use Jupyter notebooks to do so, but every time they want to try out a new idea, they need to get the notebooks running on cloud machines. This process, repeated hundreds of times a month, is time-consuming and error-prone. Moonglow reliably brings the time-to-experiment down from 5 minutes to 20 seconds. Just as Vercel and Replit abstracted away the lower levels of the computing stack for web developers and programmers, we do the same for ML researchers. https://moonglow.ai
Active Founders
Leila Clark, Co-Founder
I'm currently building Moonglow, which connects Jupyter notebooks to cloud compute. Before this, I was a software engineer at Jane Street, where I led the build-out of its equities clearinghouse connectivity infrastructure and was the technical lead of its front-office SRE team. I graduated from Princeton with highest honors in Computer Science.
Trevor Chow, Co-Founder
Co-founder of Moonglow. Prev: ML @ Stanford (Hazy Research Lab), index volatility trading @ Optiver
Company Launches
Hey everyone, we’re Trevor and Leila from Moonglow!
❌ Problem: moving experiments to cloud GPUs sucks
When you’re doing machine learning research, it’s important to try out new ideas quickly. Jupyter notebooks make that easy. But what happens when your local computer isn’t enough?
Your workflow probably looks like this:
- Go to your cluster or cloud provider
- Pick the right configs and spin up a node
- SSH into the node
- Install all the required packages
- Pull your code from GitHub
All of this is before you’ve even run a single cell in your notebook! And if you want to share your work or come back to it later, either you need to keep your GPU running (expensive) or go through this entire process again (slow).
🎉 Solution: Bring your own compute to Jupyter
Moonglow connects your local Jupyter notebooks to your cloud compute provider. With a click of a button, you can switch runtimes and scale up your experiments to the GPUs you need.
We handle all of the messy DevOps under the hood, and since your notebook lives in your local IDE, you can easily come back to it and get it running in seconds!
We currently support connecting notebooks in VS Code / Cursor to Runpod instances, and we’re expanding this to other providers soon e.g. AWS, GCP, Azure, Lambda Labs etc.
👀 Team
Trevor used to do ML research at Stanford, while Leila was a software engineer working on high-performance infrastructure at Jane Street. We started Moonglow because we’ve seen how janky and unintuitive the current tooling is for ML research, and how that is bottlenecking the pace at which researchers can validate their results at scale.
🙏 Asks
- Try out Moonglow or book a time to get set up.
- Let us know which cloud providers we should support next!
- Connect us to ML researchers you know.
We’re excited to hear from you, either at trevor@moonglow.ai or on Linkedin (Trevor, Leila).