🎉 Deepchecks raised $14m!  Click here to find out more 🚀

Not Just Testing Anymore: Unveiling Deepchecks Open-Source Monitoring Along With $14M in Funding

Introduction

As we announce our $14M in funding, we’re also going open source with AI/ML monitoring. In this post I’ll tell a bit about our journey at Deepchecks, this new milestone, and where we’re headed

At Deepchecks, we’ve made it our mission to make a dent in the way machine learning (ML) models are validated. Since the launch of Deepchecks for testing ML models in January 2022, we’ve seen an overwhelmingly positive response, with over 2,700 stars and more than 650,000 downloads. This success has made “testing ML” an integral part of the AI ecosystem. Now, we’re taking it to the next level by announcing the General Availability (GA) of our open-source monitoring solution for ML models in production, starting June 2023.

Before I dive in, here’s a sneak peek:

Sneak peek at what you get after you install. For those of you that have hands-on orientation + aren’t yet too rusty, try it by following the monitoring quickstart from our GitHub (and please star️ ️it if you like!): https://github.com/deepchecks/deepchecks

If you don’t need the background and click here to skip to the actual announcement, or here to try the open-source monitoring via our GitHub.

ChatGPT: Seeing the Power of AI Democratization with Our Own Eyes

When we founded Deepchecks, we anticipated that ML model testing would become indispensable as AI democratization progressed and as software engineers start building AI-based products without external assistance. We spoke about this with the first investors we ever met, and team members that joined our amazing team typically believed in this before they met us.

However, the recent explosion of LLM-based driven applications has publicly confirmed our belief, and to be honest — the trajectory seems faster than even we anticipated. As more people develop AI applications without being data scientists, the demand for reliable testing and validation tools becomes more and more central. Edge cases & cryptic model behavior are increasingly becoming the bottlenecks for turning cool demos into products, not just for ML people, but for mainstream software engineers as well.

Example of alleged ChatGPT bias (source)

Pioneering the Comprehensive ML Validation Suite

Our goal at Deepchecks is to enable continuous ML/AI validation for all. Our journey started with the open-source testing modules, and step by step we’re expanding this to a one-stop solution for continuous validation, encompassing testing during the research phase, monitoring, CI/CD testing, and auditing. This wouldn’t have been possible without the robust foundation established by our testing package, a project that goes back more than 1.5 years ago.

And this isn’t the only dimension in which Deepchecks is expanding. In our pursuit of a holistic offering, we’ve broadened the scope of our testing package to support various data types. While we initially focused on tabular data, we’ve since released support for computer vision (CV) and natural language processing (NLP) models, and are also building a testing solution for LLMs/GenAI (reach out if you’d like to join our beta program!).

Example of an expansion of Deepchecks: Support of testing NLP models and data, a module that was just released last week. Learn more at https://github.com/deepchecks/deepchecks

Testing. CI/CD. Monitoring.

Because ML systems are more fragile than you think. All based on our open-source core.

Our GithubInstall Open SourceBook a Demo

Ok, and Now Finally, for the Grand Reveal: Open-Source Monitoring

Our testing component is 🤗 fully open-source and geared toward the research phase, but we initially planned for our monitoring component to be closed (and optimized for companies already using Deepchecks testing). However, we discovered some significant needs for open-source monitoring within the community, that changed our perception.

Picture a group leader at a Fortune 1000 company who wants to prioritize ML monitoring but is hampered by its complexity, so it’s postponed to the next quarter. With the open-source ML monitoring component, a proactive junior ML Engineer can set up the system over the weekend and present it to the team on Monday, making a point to the team that MLOps doesn’t have to be so complicated.

Illustration of an ML Engineer that spins up an open source after being told that monitoring is important but for the next quarter 😇.

This type of need exists both for companies that want a “stay free forever solution”, but also for amazing teams with significant engineering abilities & scale, that are looking either for an interim solution when their use case is just beginning to scale, or just for the ability to set up a quick POC without needing to send out sensitive data or to vouch for a solution they haven’t even tried yet in discussions with their IT team.

To address this need, we’re offering Deepchecks monitoring as an open-source repo, that delivers a powerful and comprehensive set of features, including:

  1. Support for monitoring up to one model per deployment
  2. Root cause analysis abilities that start in the UI and continue in automatically generated Jupyter notebooks
  3. Basic user management

Our aim is to benefit the community without compromising our business model. This approach allows teams working with sensitive data to try our product without sharing their data with third parties, and easily streamlining the approval process with IT teams.

Try the open-source monitoring via our GitHub if you haven’t yet.

So What Is Our Business Model?

So given that we’re beginning to release open-source repos with various parts of our core logic, you may be wondering, what’s the catch? What’s our business model?

The answer is pretty straightforward: It’s a classic open-core strategy. There are a few groups of features that aren’t part of the open-source, such as advanced security/identity management, more scalable deployment options, a centralized dashboard for many models, and audit/compliance templates. A subset of these features is important for the teams we’d like to sell to, but the open-source is still powerful enough without them to be attractive for most teams.

We have quite a few interesting modules in our roadmap. One of the more interesting modules we’re building is dedicated support for testing & evaluating LLMs. If this field is interesting for you I recommend you join LLMOps.space, a global community we’re beginning to put together for LLM practitioners. See you there!

In the meantime, if you’d like to read similar stuff in the future, follow me on LinkedIn or subscribe to Deepchecks’ newsletter.

Testing. CI/CD. Monitoring.

Because ML systems are more fragile than you think. All based on our open-source core.

Our GithubInstall Open SourceBook a Demo

Recent Blog Posts

Training Custom Large Language Models
Training Custom Large Language Models
How to Train Generative AI Models
How to Train Generative AI Models