My name is Kasra, and I am a Senior Applied Scientist in the Zalando Assistant (ZA) team. My current work focuses on exploring applications of Large Language Models (LLMs) and multi-modal models in the development of AI assistants.
I have been actively involved in assessing potential risks associated with LLM-integrated applications and devising strategies to mitigate these risks. Recently, The Wall Street Journal highlighted our collaboration with the Information Security team at Zalando to enhance the safety and security of these emerging technologies.
The Role of the Zalando Assistant (ZA) Team
The ZA team is a part of the PRSB (Personalization, Recommendations, Search, and Browse) division. Our team leverages large datasets, data science techniques and AI to help customers find relevant and personalised products.
Specifically, we are developing the Zalando Assistant to empower customers to interact with Zalando in a more intuitive way. Through leveraging recent developments in LLMs and Multimodal models, we enable them to navigate our extensive assortment in their own words, simplify product discovery and improve the overall shopping experience.
Presenting at UX Nordic: Safeguarding AI
I recently had the opportunity to co-host a workshop at the UX Nordic Conference, where we explored the topic of AI safety. Our workshop focused on how product designers, user researchers, and AI practitioners can collaborate to ensure AI-powered products are developed and deployed responsibly. We introduced several frameworks, such as Consequence Scanning, the Abusability Testing Framework, and the Risk Matrix, which teams used to assess and mitigate the risks associated with different AI-driven products.
Participants were divided into teams, using these frameworks to evaluate the consequences of example products, prioritise identified risks, and develop mitigation strategies. At the end, each team presented their findings and proposed how to implement effective safety measures. Our discussions spanned a range of topics, from the intended or unintended consequences of driver-assistance systems to those of personalised health assistants.
Relevance to My Work
The workshop was inspired by our work on the Zalando Assistant. Before launching in August 2023, we assessed the intended and unintended consequences of AI-powered assistants in e-commerce.
In particular, we explored the ethical, social, and security risks posed by LLMs and developed strategies to mitigate them. Given the evolving nature of AI safety, hosting this workshop allowed us to share our insights with the broader community, contributing to a wider conversation about responsible AI usage.
The Power of Collaboration
At Zalando, collaboration is at the core of our work, and we collaborate across job families and roles every day. Building responsible AI solutions requires input from various experts, and so it’s important that we collaborate effectively whilst building and deploying products.
Working with a diverse team of specialists has been key to mitigating the risks associated with advanced and evolving AI models and the systems that integrate them. Our shared commitment to a safe and ethical AI deployment is only made possible through unique ideas and perspectives, which are crucial for effectively tackling the challenges of protecting AI systems.
Effective Collaboration Across Teams
Effective collaboration requires a multifaceted approach, and based on my experience, several strategies can be particularly helpful.
First, establishing a common language is important, which means making a conscious effort to avoid role specific jargon. Whilst this can be challenging, it’s important we communicate in a way which everyone can understand. This could also involve hosting talks or workshops for colleagues in other roles, to ensure they understand the topics at hand.
Additionally, strong communication skills play a key role. This can mean clearly articulating ideas or maintaining clear and detailed documentation. This allows colleagues from other job families to review materials on their own schedule and provide feedback when needed. I enjoy writing and documenting, and my time at Zalando has given me valuable opportunities to refine these skills.
Lastly, being open to new ideas and methods is crucial. In multidisciplinary settings, there are often moments when you may not fully grasp the reasoning behind certain approaches. Rather than dismiss these ideas, I’ve found it beneficial to engage with my colleagues and explore their thought processes.This not only fosters mutual respect but often uncovers valuable insights that might otherwise be missed.
Living the Founding Mindset (OFM)
I have several favourites, but if I had to choose just one, it would be “Fly high and dive deep”. This OFM emphasises the importance of maintaining a broad perspective while being ready to delve into details when necessary, without getting stuck. I like this OFM because it encourages deep thinking about what truly matters and helps in allocating resources effectively. I use this OFM daily.
