Alexander Machado from appliedAI Initiative

In this episode, host Rod Rivera interviews Alexander Machado, the Head of Trustworthy AI for the Applied AI Initiative in Munich, Germany. With over a decade of experience in artificial intelligence, Alexander shares valuable insights into the evolving landscape of AI engineering. The conversation covers a wide range of topics, from the fundamentals of Machine Learning Operations (MLOps) to the emerging field of Trustworthy AI. Alexander discusses the challenges faced by companies implementing AI projects and offers practical advice on structuring AI teams and managing projects effectively. Listeners will gain a deeper understanding of the AI project lifecycle, the importance of data engineering in AI initiatives, and the impact of new regulations like the European Union's AI Act. Alexander also shares his perspective on the shift towards cloud-based solutions and the future of AI in 2024. Throughout the interview, Alexander provides encouragement to Latin American engineers, emphasizing that they are equally capable of succeeding in the global AI industry. His journey from Venezuela to Germany serves as an inspiring example for aspiring AI professionals. This episode is a must-listen for data scientists, AI engineers, and anyone interested in the current state and future direction of artificial intelligence, particularly in the context of European regulations and global industry trends.

Takeaways

• MLOps ensures AI models maintain high performance in production
• Open-source tools for MLOps: Lakehouse, MLflow, Evidently AI, Seldon Core
• AI is a competitive advantage for businesses
• Product management for ML differs from traditional approaches
• Cloud migration is a growing trend in AI
• Trustworthy AI focuses on creating fair, non-discriminatory models
• The European AI Act will impact AI development and deployment

💡 Top Takeaway: MLOps and Trustworthy AI are crucial for successful AI implementation, especially with upcoming regulations.

🔗 Resources mentioned

• MLOps Workbook by Applied AI (free course)
• European AI Act

Episode Transcript

Introduction and Professional Background

Rod Rivera: Today, I have the pleasure of being with Alexander Machado, Head of Trustworthy AI for the Applied AI Initiative in Munich, Germany. Welcome, Alexander.

Alexander Machado: Thank you very much, Rodrigo. It's a pleasure to be here with you.

Rod Rivera: Looking at your resume, I see you've had a long career. How did you get to where you are now? Tell us a bit about your journey.

Alexander Machado: Thank you for the question. I've been working in Artificial Intelligence for over 10 years. It all started during my master's degree at the Technical University of Munich. My first internship was at BMW in the Data Science area, where I worked on video compression using AI and vehicle sensor information.

That experience was eye-opening for me. I did very well, publishing two conference papers and obtaining a patent. From that moment on, I decided to dedicate myself to Data Science, shifting my previous focus from telecommunications and electronic engineering.

After university, my first job was at the Max Planck Institute, the largest basic research center in Germany. There, I worked as a data scientist, managing the Data Pipeline of the Max Planck Digital Library.

Then I joined Applied AI as a senior artificial intelligence engineer. MLOps became a crucial topic, and I was promoted to Head of Machine Learning Operations. Starting January 2024, I'm taking on a new role as Head of Trustworthy AI, focusing on how to combine MLOps and best practices with the European AI Act.

MLOps: Definition and Importance

Rod Rivera: What is MLOps and how would you describe it in a few words?

Alexander Machado: MLOps, or Machine Learning Operations, is about ensuring that a model developed in the laboratory can be deployed in production systems while maintaining high performance. Unlike traditional software, AI models are not static. Their performance can degrade over time due to changes in data distribution. MLOps is responsible for automating and monitoring these models throughout the entire Machine Learning lifecycle.

Getting Started with MLOps

Rod Rivera: If I'm a data scientist and want to start with MLOps, how can I do it? Are there essential tools or a methodology I should follow?

Alexander Machado: Definitely. First, I would recommend taking an MLOps course. At Applied AI, we recently developed the "MLOps Workbook," which explains MLOps principles, the Machine Learning lifecycle, and how to plan and execute AI projects. This course is available for free on our website.

As for tools, we use several open-source ones:

• For data versioning: Lakehouse or Delta Lake
• For experiment tracking: MLflow or GearML
• For model monitoring: Evidently AI
• For deployment: Seldon Core or RaySurf

Rod Rivera: Are these tools easy to connect with each other?

Alexander Machado: There are several approaches. Many companies are migrating to the cloud, where platforms like Azure offer these tools already integrated. For companies that can't use the cloud due to privacy reasons, there are on-premise platforms like DataIQ or Domino Data Labs. It's also possible to create your own platform by integrating open-source tools, although this can be more complex.

Implementing AI Projects in Companies

Rod Rivera: How do you convince companies to implement data science and machine learning projects?

Alexander Machado: The key is to show that AI is a competitive advantage. If a company doesn't adopt it, their competitors will, and they'll be able to do in a month what takes others three. Generally, we start with a prototype, then move to production, and when they see the value, they're more willing to implement more use cases.

Product Management for Machine Learning

Rod Rivera: Are there differences between traditional product management and machine learning product management?

Alexander Machado: Yes, there are significant differences. In AI projects, there's much more initial uncertainty and risk. It's difficult to accurately estimate how long an AI project will take. That's why at Applied AI we use a four-phase model:

1. Planning phase: Brainstorming use cases and business value analysis.
2. Exploration phase: In-depth analysis of available data.
3. Implementation phase: Model development to achieve objectives.
4. Production phase: MLOps integration and model deployment.

This approach helps us reduce uncertainty and validate the feasibility of AI projects.

Evolution of the AI Industry

Rod Rivera: What has changed in the AI industry in recent years?

Alexander Machado: A clear trend is the migration to the cloud. Previously, many companies wanted to create their own AI platforms with MLOps, but now they're increasingly moving towards cloud solutions. We've also seen an increase in demand for generative AI use cases, although it's important to remember that it doesn't solve all problems and traditional AI remains crucial for many use cases.

Trustworthy AI and European Regulations

Rod Rivera: Tell us about your new role in Trustworthy AI and how it relates to new European regulations.

Alexander Machado: Trustworthy AI refers to creating AI models that are fair and non-discriminatory. With the upcoming implementation of the AI Act in Europe, companies will need to ensure that their AI models meet certain requirements, especially for high-risk use cases. This involves implementing MLOps practices and ensuring the transparency and fairness of models.

Final Advice for Latin Americans in Tech

Alexander Machado: An important piece of advice for Latin American listeners: Don't underestimate your capabilities. My experience has shown me that engineers in Latin America are just as capable as those anywhere in the world. If you want to come to Europe or Germany, apply with confidence. Opportunities are there for those who seek them.

Rod Rivera: Thank you very much, Alexander. It's been a pleasure having you today on Ingeniero AI.

Alexander Machado: Thank you, Rod. It's been a pleasure.