Last summer we first unveiled Spell with the mission of expanding access to Deep Learning and AI. The incredible advances in AI coming out of research were open-source and freely available, but getting them running on the right hardware was a challenge preventing more people from making use of them. So we built Spell to be the simplest way to get started from anywhere.
Since then, tens of thousands of experiments have been run using Spell and we’ve heard from engineers, researchers and students that our tools are the primary reason they’re able to do work in AI.
But we also heard from companies who told us the tools available for building real products with AI were immature, fragmented and costing them time. So we decided to make the next version of Spell a complete end-to-end system for exploring, training, building, automating and serving models built with deep learning.
And that’s what we’re launching today — a new look, a completely revamped web console, and a suite of new features designed to power AI and ML for teams and companies.
Our new look
First, we are proud to launch our redesigned website and console, which went live earlier today. You’ll find a completely new experience on spell.ml, including a new brand identity and an updated style. The new site is better laid out, more accessible and now fully responsive — which means you can keep track of your runs on desktop, mobile, and tablet.
Run reports have also been redesigned to be easier to read and to highlight the results of your experiment right at the top. And you’ll find dozens of other usability and performance improvements throughout the site.
Today we’re also launching Organizations. Now everyone on a team can see each others’ experiments, results, and comments in one convenient place. Admins can get alerts on the status of runs across an organization, making it easier to keep track of the overall progress of your experiments.
Taking a trained model and deploying it as an API that can be used in production can take a lot of boilerplate and unnecessary engineering effort. With Spell a single command takes your trained model and deploys it as a secure and scalable API.
Spell uses Kubernetes to automatically scale models as necessary, and we provide metrics, monitoring and logs for everything running in real time.
When we launched our hyperparameter search feature in the fall it was an instant success. Developers were able to quickly run parallel searches to optimize models.
Now with our Hyperparameter visualizer, you can kick off a search with one command and visualize the performance of each experiment in real time.
Organizations can now deploy dedicated clusters, keeping all data, experimentation and model serving within your own AWS or GCP infrastructure. The benefits of running in your own cluster include:
- Privacy and security: Data never moves out of your own network/cloud
- Billed by use: We charge based on peak concurrent instances, so you’re never overpaying for compute power.
When operating behind your own firewall Spell gives you simple tools to manage your deployment and keep track of how much compute your team is using. Manually adding machines and data results in wasted resources and insecure processes, so we handle machine allocation automatically so you’re never overpaying or wasting resources.
We’re also happy to announce that we’ve closed $15M in new funding from Eclipse Ventures and Two Sigma Ventures. We’ll be using this new capital to add features, power more and larger organizations, and keep making AI and ML more accessible to developers and companies everywhere.
If you’re interested in adding AI and deep learning to your company reach out to us at firstname.lastname@example.org! Or if you’re an engineer or student, sign up at web.spell.ml/register for free.
Also, we’re hiring!
We’re barely two weeks into 2019 and it’s already clear this is going to be an incredible year at Spell!