Domino's Enterprise AI Platform

My main use case for Domino Data Science Platform varies due to the nature of this large bank, the leading bank in Europe, where we have numerous data science and machine learning models that need to go into production for many customer and internal employee applications for prediction, particularly in the insurance domain for predicting fraudulent cases, checking customer retention for insurance renewals, and providing financial recommendations. Therefore, the main use cases are primarily bank-related, and my role focuses on the deployment side of these models.

Recently, I deployed a model for an email triaging use case using Domino Data Science Platform, which is specifically designed for machine learning model deployment, unlike other cloud providers that may not offer all services in one platform. For versioning code, I used GitLab, and for DevSecOps, I applied tools like SonarQube, Fortify, and Nexss to scan for vulnerabilities, enabling code scans. I then utilized Domino Data Science Platform for the continuous delivery part by leveraging its compute capabilities such as GPUs and CPUs to run automated jobs from Git for model deployment and experiment tracking used by data scientists. Once the best model is finalized, it gets deployed within Domino Data Science Platform's model registry, which is converted to endpoints for application consumption. We monitor drift with Domino Model Monitoring, and since Domino Data Science Platform lacks a built-in resource monitoring system, we utilize external tools such as Grafana and Prometheus to track CPU and other resource usage during this activity.

Domino Data Science Platform has positively impacted our organization, especially since we adopted it within this bank from the beginning. There is a strong understanding of how we utilize Domino Data Science Platform, along with constant support from the Domino Data Lab team, as we participate in monthly workshops where they present the latest features. We discuss our challenges, and they offer effective solutions, leading to very positive feedback from my colleagues about the platform, ensuring that we will continue using it without considering alternative tools for model deployment.

The outcomes we have experienced with Domino Data Science Platform have been significant, with reductions in time thanks to its flexibility, which allows us to manage resources according to our needs and project requirements.

In my opinion, the best features Domino Data Science Platform offers relate to model management, specifically the model registry and the environment flexibility which allows us to create our own environments and define base images. We can write Docker instructions for these environments, making them suitable for model APIs or applications, while also ensuring that each revision we create is beneficial for audit and governance purposes. The entire process, including model APIs, endpoints, environments, and projects, is versioned, which is crucial for audit compliance and governance.

The features of Domino Data Science Platform have facilitated a smoother workflow by allowing us to automate the deployment of environments that serve multiple use cases. We utilize a unique method for installing packages, which is an advanced strategy leading to quicker and more reliable deployments. This approach involves using environment revisions so that when changes occur, we update these environments while maintaining specific revisions for various endpoints and applications, enabling us to automate processes around which model and endpoint use which specific revision, thus saving costs and resources while simplifying our operations. This level of functionality is truly advantageous in Domino Data Science Platform.

While I believe I have covered most areas, I want to emphasize the security features in Domino Data Science Platform; as an admin, I manage access permissions to projects, defining who can view, use, or edit applications. The flexibility to set security levels for each service is a key feature that I appreciate.

Improvement areas for Domino Data Science Platform could relate to resource monitoring capabilities; adding visuals that stakeholders can review would enhance awareness of resource usage and its impact on applications and costs. Additionally, while we have Domino Model Monitoring, I would like to see more model monitoring options being developed to benefit us and data scientists.

As an MLOps engineer, my primary concern focuses on continuous training and continuous integration and deployment, and I believe Domino Data Science Platform could enhance its services for monitoring capabilities. Specifically, improving drift reduction by analyzing ground truth data alongside model metrics would be beneficial, as would include automatic triggers for deployments within Domino Data Science Platform instead of relying on external tools such as Git.

For the past year, I have been using Domino Data Science Platform for model deployments, taking models to production, and helping data scientists from the workspace until the endpoints.

In terms of my experience deploying models using Domino Data Science Platform, I have previously worked with Google Cloud and Vertex AI, as well as Azure Machine Learning Studio, but I find that Domino Data Lab is a much more advanced platform specifically tailored for model deployment, which significantly eases our workflow. It offers flexibility to run automated jobs, automated CI/CD flows, and optimally utilize resources, allowing me to identify Domino Data Lab as an excellent and advanced system for model deployment. Their model monitoring is impressive, and with their latest version introducing AI and AI agents, I see future potential, although we currently do not have these features due to our bank's strict security policies. Importantly, Domino Data Lab stands out in terms of security compared to other cloud providers such as Azure, Google, or AWS, making it suitable for on-premises use without concerns about data leakage.

I rate Domino Data Science Platform a 10 out of 10 because of the numerous positives I have mentioned and the fact that I cannot think of any replacement other than Domino Data Science Platform. Domino Data Science Platform is continually evolving, and I look forward to the upcoming version that will feature enhancements such as AI and AI agents, which I believe will further improve our experience.

Regarding the AI capabilities of Domino Data Science Platform, we had a workshop discussing AI governance and security, although I cannot comment extensively on it since we do not have those AI features and agents currently available in our bank.

Domino Data Science Platform is deployed on-premises within our organization, specifically on IBM DNZR data systems, as we prioritize security and therefore do not utilize public cloud services.

My advice for others looking to use Domino Data Science Platform is to leverage its full potential by understanding each layer of the deployment process, from model creation to project design, and ensuring clarity on where the code operates, its versioning, the environment structure, hardware resources available, and the applicable security levels for every service. I rate Domino Data Science Platform a 10 out of 10.

We used Domino Data Science Platform for developing and working with machine learning models. It facilitated end-to-end development processes. Domino is based on Git, enabling collaboration similar to using Git. Each user operates on their own equivalent of a branch or fork, and once finished, they can merge their changes with the main project.

Domino made developments easier, which indicates a good investment as it reduced development time significantly.

The workspaces, which are like wrappers of Docker containers, made it easy to start development environments using Domino. Additionally, deploying a model using Domino was quick. Domino's API also speeds up the deployment process.

The deployment of large language models (LLMs) could be improved. Currently, Domino provides a simple server that cannot handle big deployments, which is not suitable for LLMs.

I have been working with Domino Data Science Platform for four years.

Domino is quite stable. I would rate it nine out of ten for stability.

Since we didn't work with big data, stability and scalability were not major concerns. Domino is based on Kubernetes, allowing easy configuration of the number of workers or machines needed, with no issues in scaling.

The technical support team was very helpful. They sometimes visited us in person. However, creating new features could take some time because they had to consult other customers.

A team of six or seven people were needed for maintenance on Domino.

Domino delivered return on investment by facilitating easier development work, even though I don't have specific numbers.

There is a licensing fee per user and additional costs for customer support. I don't know the exact pricing, but it's likely above average.

It's important to have a DevOps team well-versed with cloud-native solutions to manage Domino effectively. Relying solely on data scientists might not be sufficient.

I'd rate the solution eight out of ten.