March 31, 2022
7 min read

Why & how we decided to make Giskard Open-Source

We explain why the Giskard team decided to go Open-Source, how we launched our first version, and what's next for our Community.

Open Ocean
Alex Combessie
Open Ocean
Open Ocean

"Captain's Log, Stardate 2022.3. We have entered a new phase of our journey in the MLOps sector of the AI/ML universe. A phase where companies build in the open and new communities form. Our mission is one of collaboration, performance, and balance for ML models.”

First, our thoughts go to the people of Ukraine and Russia. We are very saddened by the situation and wish for peace.

The Giskard team has been active & growing this month. First, we onboarded our first employee,  AI Evangelist, and master storyteller. He will help us build a community of passionate people, who want to tackle the challenges of ensuring the quality of ML models.

The key milestone for us has been the launch of our Open-Source offering. This was a big decision for us, and we worked hard to make it a reality.

In this post, I will explain why we came to realize that Open-Source was the right way for us, and share our experience on how we launched it.

⚖️ Why go Open-Source? Pros & Cons

The #1 pro factor was transparency. It is one of our core values as founders. We started Giskard because we hate black-box AI. We felt this value had to be embedded not only in our company culture but also in our product. How can one fix the issues of transparency in AI by providing closed-source software, i.e., another black box?

Second, to build the best Quality Assurance platform for AI, we need community engagement. Open-Source is all about empowering people to give feedback and contribute to the product they use. Customization to specific needs is key to building the best product. The more people contribute, the more we benefit from this network effect, improve our features, and ship faster.

That’s why there has been a massive shift toward Open-Source in the AI community in the past 3 years. Researchers make their papers publicly available on Arxiv and push their models on Huggingface. Data scientists use open-source libraries to train models. ML engineers adopt open-source MLOps tools and follow best practices from the software engineering industry, inspired by DevOps.

To confirm this, we asked our community on LinkedIn:

Survey on the importance of Open-Source for ML tools
Survey on Machine Learning tools

You asked for it - and we delivered!

Lastly, we evaluated the answer to give to the pragmatic “Con” question: can we build a successful & sustainable company based on an Open-Source Software (OSS) product? In other words, what is the path to revenue? To answer this question, we spent a lot of time gathering advice from some of the leaders in the OSS world: Huggingface, Strapi, Quickwit to name a few.

We realized that yes, it is possible. Here is how we got started!

🚀 How we launched our Open-Source offering

Licensing 📝

The very first thing to choose when you go Open-Source is the software license. There are multiple choices with different implications. Here are some of the key points about the licenses that we had on our shortlist:

  • MIT - Permissive, short but does not cover explicitly patents rights and trademark restrictions
  • BSD (2 or 3 clause) - Permissive, compatible with proprietary or open-source licenses, does not grant any patent rights
  • AGPL - Copyleft, fewer people would be able to use Giskard unless their product is also open-source, which may lead to fewer contributors in the future
  • Apache 2 - Permissive, more explicit than MIT, contributors provide an express grant of patent rights, recommended by GNU and Google

We decided to start with the Apache 2 license.

Packaging 📦

We had originally designed our software with a SaaS deployment in mind. We started by removing cloud-specific information from our Github repository. Then, we optimized our build system to make sure the software could easily be installed and hosted on any environment using Docker.

Now, Giskard can be installed in 3 lines of code, in under 3 minutes.

Documentation 📖

When you go Open Source, it’s crucial to have clear and comprehensive documentation. It also needs to be collaborative, so that the community is able to easily complete it. This is why we chose Gitbook as our documentation tool. We really liked how Gitbook is integrated with Git to do version control between multiple contributors.

You can get started on: https://docs.giskard.ai/start/.

Giskard Documentation
Get started with Giskard

Community 💬

Our vision for the Giskard Community is "Made by ML engineers, for ML engineers”. We aim to attract new contributors by providing fast & transparent information and support. We want to facilitate open communication where community members can give us direct feedback and mutually help each other.

We ran a community survey on LinkedIn, which pointed out Discord as the best platform. It is increasingly popular among open-source communities and offers very good animation & moderation features at scale. To help structure the discussions, we create specific channels for specific goals: announcements, product ideas, support, etc.

Star us on GitHub and come onboard the Giskard Community!

Giskard Community on Discord
Our Discord Community

📍 What's next?

We are actively developing the next module of our product: AI Test. It will help Data Scientists and ML Engineers build and execute complete test suites on ML models in no time. The goal is to ensure the performance & balance of ML models before deployment. The first version will land in a few weeks. Contact us if you would like to be one of our design partners!

We are also working with a designer on a completely new brand identity. We will keep the Giskard name for now, but change our logo to… an animal. If you guess which animal, let us know in the comments. We will ship an exclusive reward to the winners!

We hope you found this post useful and interesting. Moving forward, we will publish this kind of public report on a monthly basis. We are very interested in your feedback to improve the next one.

If you like what you read, please subscribe and reshare it!