Charting new heights: Keyvan Sadri from Lufthansa Group Digital Hangar on AI's role in aviation's current development and its future

The aviation industry has soared on the wings of technological advancements for decades, but recent AI breakthroughs are set to turbocharge its evolution. Airlines are now harnessing artificial intelligence to streamline flight planning and operations, reducing congestion and carbon footprint along the way. In 2023, the number of passengers carried by air in the EU almost reached the pre-pandemic levels, bringing back the need for efficient mass solutions.

But something else changed on the way - we now have broad access to powerful large language models. What does it mean for the industry? In a discussion with Jerzy Jarzębowski, Keyvan Sadri reveals the sky-high potential of AI in aviation, drawing from his hands-on experience with projects for Lufthansa. Buckle up and explore his insights into how AI is reshaping the future of flight!

Key Takeaways from the Conversation

Board up the AI journey with a solid data system: Recognizing data science needs is crucial for successful AI implementation. To implement a model, you should develop a solid data architecture to easily scale your project in the future. Think about it beforehand, especially if you need to include external data sources. Key points to consider? Your system should be fast, sustainable, and technologically up-to-date.

The sky’s the limit… or so can be your model: Whether you choose a large or small language model, there will be risks involved. Small models require a lot of work and attention to prevent limited context and legal implications. On the other hand, large language models may lack transparency and create dependencies on external providers. In the end, it all comes down to weighing the pros and cons carefully.

Data management is what you need: Maintaining a proper data warehouse will give you a solid foundation for creating and maintaining an AI solution. Data visualization and fast information extraction will help you make the best use of the data available. Once you have this resolved, you’ve prepared the ground for advanced analytics and prediction systems to step in.

Conversation with Keyvan Sadri

Jarek Jarzębowski: Let’s start with a bit of background. Can you tell us a little bit more about yourself and your expertise?

Keyvan Sadri: Sure! My name is Keyvan Sadri. I am originally a physicist and did my Ph.D. at the Max Planck Institute in Heidelberg. After finishing my Ph.D., I started working as a data science consultant in the energy industry, financial services, retail, and manufacturing. I’ve done several projects, and my last consulting job was at KPMG, where I was a consultant for four years. Then, I joined Lufthansa as a senior manager in data science, focusing on developing and delivering fraud AI and advanced analytics products related to our customer domain.

Jarek Jarzębowski: Can you briefly tell us a bit more about the product you are working on?

Keyvan Sadri: Lufthansa, which I work for at the moment, has started an ecosystem called the Digital Hangar. The aim was to develop customer-centric products, for example in customer service, the travel experience, and our B2B clients. 

I am the Product Owner of the AI team, where we develop AI products for various parts of this environment. We have products that identify personal information and documents to enhance our GDPR compliance efforts. Moreover, we have also delivered the Claim Summarizer which is our first Gen AI powered product. It can extract facts from customer claims.

Jarek Jarzębowski: Great! It’s a great variety of products with different use cases. I’m also wondering if you could give us a short explanation of your point of view on the landscape of AI in logistics and airlines in general. What is the current state of AI, and what advancements have been made in recent months or years?

Keyvan Sadri: Let me start with the challenges in the airline, especially the passenger airline industry. After the pandemic, many countries removed travel restrictions, which made the demand for flights surge. At the same time, there were many operational problems. For example, many airlines had parked their aircraft and needed to perform maintenance to ensure they were ready to fly. Additionally, many people had left the industry. As a result, the airlines had to hire new staff, including cabin crew, amidst high demand.

These operational problems put significant pressure on customer service. Customers had to rebook flights, call customer service centers, and locate lost baggage, which are all parts of the customer journey. To address these issues, Lufthansa started the Digital Hangar to improve parts of the customer journey by offering digital products. Digitalization, including AI, can help enhance customer service, rebooking processes, claim management, and even meal selection before flights.

Jarek Jarzębowski: Are you using AI and technology for demand forecasting, such as predicting meal requirements or other aspects?

Keyvan Sadri: Yes. Demand prediction is a well-known use case in network planning. Like many other airlines, Lufthansa uses machine learning to enhance the demand prediction.

Jarek Jarzębowski: Can you share some challenges you have encountered while working on these AI projects?

Keyvan Sadri: One major challenge is GDPR compliance, but we handle it positively, ensuring clear guidelines on what we can and cannot do with data. Another challenge is the scalability of AI solutions. When a model goes into production, it must consistently deliver results, whether for decision support or automation. Many companies develop models that never reach production, leading to wasted investment. Lufthansa is well-positioned to produce AI models, but it still requires time, money, and standardized processes.

Jarek Jarzębowski: What is the biggest challenge in putting a model into production?

Keyvan Sadri: The main challenge is companies often do not recognize data science needs its own environment and ecosystem. Data science involves understanding the business need, finding the right data, having proper data engineering and architecture, and establishing a sustainable MLOps process. Many initiatives start with hiring data scientists without considering the broader ecosystem needed for success. Companies must view data as an ecosystem to realize value from AI.

Jarek Jarzębowski: What best practices should companies follow for data management and AI implementation?

Keyvan Sadri: Based on my consulting experience, using data involves three steps: data management, business intelligence, and advanced analytics and machine learning. First, companies should manage data effectively, collecting it from all sources and maintaining a proper data warehouse. Next, they should use business intelligence to understand their data through visualization and fast information extraction. Finally, they can move to advanced analytics and machine learning for prediction and decision support.

Jarek Jarzębowski: Should companies know their end goal when starting with data management, or can they apply best practices and use the data later?

Keyvan Sadri: In the first step, companies should at least know their own data and collect it correctly, ensuring a sustainable system that is technologically up-to-date and fast. While they might not need specific data science use cases initially, having a good data architecture will support future data science projects. External data sources will also be necessary for advanced analytics, and a well-designed data ecosystem can handle this.

Jarek Jarzębowski: What do you think about the future of AI in aviation and logistics? How should companies prepare for it?

Keyvan Sadri: The future holds many possibilities, such as updating prediction methods with more data and advanced techniques, using the Internet of Things for maintenance, and leveraging generative AI for text and voice processing. Companies should be careful with the cost and implementation of generative AI, ensuring defined business value. Smaller language models can be useful for specific tasks, reducing dependency on expensive external providers. Preparing involves focusing on defined use cases and exploring small language models.

Jarek Jarzębowski: Are there any risks with using large language models versus smaller, internal ones?

Keyvan Sadri: Large language models can be expensive and create dependencies on external providers. Training small language models is a significant investment, requiring proper deployment and monitoring. Risks include hallucination or wrong information, which can have legal implications. Large language models lack transparency, making it difficult to understand their decision-making process. Companies should be cautious about these risks and ensure robust monitoring and validation.

Jarek Jarzębowski: Apart from data management and MLOps, what else should companies focus on when introducing data science and AI?

Keyvan Sadri: Companies should balance having strong business cases with enabling innovation. Over-focusing on business value can stifle innovation. Companies should find ways to encourage innovation, such as dedicating time for experimentation and innovation sprints. In the Digital Hangar, we use a SAFE framework with quarterly innovation sprints to foster creativity and new ideas.

Jarek Jarzębowski: Can you share some successes from innovation sprints or hackathons?

Keyvan Sadri: We’ve seen significant success from innovation sprints and hackathons. These events bring together people from different domains within Lufthansa to work on new ideas. While I can’t share specific product details, I can say that these initiatives have led to valuable insights and the development of impactful products.

Jarek Jarzębowski: Thank you for sharing your insights and experiences, Keyvan. This has been a very enlightening conversation.

Keyvan Sadri: Thank you, Jarek. It was a pleasure discussing these topics with you.

Keyvan Sadri’s Background

Keyvan Sadri is an expert in AI and data science with a Ph.D. in Theoretical Chemistry from Universität Heidelberg and a Master’s in Physical Chemistry from Sharif University of Technology. His career spans multiple industries, including roles as a Senior Data Scientist at KPMG Deutschland and a consultant for BearingPoint and Publicis Sapient, where he led various big data and AI projects.

Currently, Keyvan collaborates with Lufthansa, serving as the AI Product Owner and Interim Lead of the Community of Practice for Data Science and Machine Learning. His work focuses on operationalizing AI models, optimizing flight operations, and enhancing customer service through AI innovations, ensuring the aviation industry remains at the forefront of technological advancements.

About Lufthansa Group Digital Hangar

The Lufthansa Group Digital Hangar aims to make flying easier and more enjoyable by combining physical, human, and digital experiences across all Lufthansa Group airlines. The project focuses on digitalizing every part of your journey, from booking and check-in to in-flight services and baggage handling. Working on creating better online shopping and booking experiences, they drive personalization efforts to ensure you get the right information and offers at the right time. Their goal? Leading the way in digital solutions for the airline industry.

Closing Thoughts

The aviation industry is regaining strength after pandemic-induced setbacks. However, this means not only business opportunities but also challenges. Especially in the era of climate change, airlines must focus on data-driven solutions that continuously optimize their operations in terms of fuel consumption, waste generation, and overall efficiency. 

At the same time, passenger expectations are rising, and airports must handle an increasing number of airlines and connections. This, in turn, leads to delays, impacting customer satisfaction. A data-driven approach involving advanced AI predictions can help airlines manage fluctuating demand and meet these expectations. 

Will the aviation industry reach new heights with expanding AI capabilities? Everything suggests that the coming years could be game-changing. Let’s wait and see what Lufthansa has been working on!