Fall Campus Opening Ceremony

Placeholder for Jan: Talk 1

• Jan Janke (CERN)

to-do

Introduction to Alarming in Industrial Controls sector

• Matthias Braeger (CERN)

Alarming in the industrial controls sector plays a critical role in ensuring the safety, efficiency, and reliability of automated systems. This process involves monitoring key parameters of industrial equipment and processes, generating alerts when predefined thresholds are breached, and enabling timely intervention to prevent malfunctions or safety incidents. Alarming systems are designed to provide real-time notifications to operators, facilitating quick decision-making and mitigating risks. An effective alarming strategy reduces downtime, enhances process control, and maintains compliance with safety regulations, making it a cornerstone of modern industrial automation and process control systems. This lecture will provide you a good overview about Industrial Controls strategies for alarming, based on real examples from CERN's Technical Infrastructure Operators in the CERN Controls Centre (CCC).

QuantumConnect: Particle Collision Challenge

• Cristian Schuszter (CERN)

Step into the world of quantum coding with QuantumConnect, where you will craft your own Java programs to compete in a Connect 4-style game tournament. Like quantum particles racing through a collider, your program will need to react swiftly and precisely to outmaneuver opponents. Participants will have 45 minutes to create and submit their solutions, which will then go head-to-head in strategic battles, testing their adaptability and strategy. In this fast-paced, quantum-inspired competition, outcomes will be as unpredictable as particle behavior at the quantum level. Competitors must rely on logic, quick thinking, and efficient coding to claim the title of QuantumConnect Champion. Join us for a session of intense collisions and high-energy gameplay, where coding and quantum mechanics collide in the ultimate challenge!

Placeholder: Lecture Given by Tartu University

• Karl Kruusamäe

Modernizing Legacy Code: A Case Study of React Implementation at CERN.

• Dmitry Kekelidze (CERN)

Have you ever been knee-deep in JavaScript, trying to spice up your web-site or web application, and things start getting insanely complex? Well, we hit that wall at CERN too with one of our flagshit applications, and guess what? We tackled it! Come along and join me for this talk, where I will show you how we used React to build a new framework to modernize our code that's been around for more than 20 years. Whether you're a beginner trying to navigate your way, or a veteran, there's something for everyone.

Power BI from A to Z: What It Takes to Make It Work

• Christina Tsiplaki Spiliopoulou (CERN)

Power BI seems easy: connect to some data, build a few visuals, publish a report, and suddenly everything looks beautifully data-driven. Then real life happens. When starting a new project, you are not just building visuals. You are juggling different business needs and the many options Power BI offers. In this session, I will take a practical tour of Power BI from A to Z, using my own experience to share what I wish I had known earlier. We will look at the full journey behind a reporting solution, from data sources and modeling to report design, publishing, governance, and adoption. We will also touch on the things that make reports work in the real world: building models that perform well, designing reports people can actually understand, and making choices that still make sense once a report is used by more than one team. Anyone can build a dashboard. The real challenge starts when real users, real data, and real ownership enter the picture.

Intelligent Data Systems and AI for Autonomous Space Technology (Part 1)

• Sudeshna Sengupta

Space technology has always been a fascinating topic that encompasses various fields of study. The current major space mission, Artemis II, to fly by the moon, has gripped the world’s attention. However, the amazing advancements in the field come with their own set of challenges now. Space has become congested and resource-constrained, which has now made the need for efficient and sustainable spacecraft operations more critical than ever. Most traditional space missions rely heavily on transmitting large volumes of raw data to Earth for processing, which requires extensive resources and is inefficient. But recent advances in onboard computing and artificial intelligence are reshaping the way such missions can process, filter and interpret data directly in orbit. Missions such as ESA’s ɸ-Sat series and OPS-SAT show how machine learning algorithms can identify relevant data in orbit, such as anomalous space weather and disruptions from space debris, which significantly reduces unnecessary data transmission. In doing so, these systems can lower power consumption, optimise bandwidth usage, and enable smaller satellites to perform complex tasks. This topic is a two-part lecture which explores the growing role of intelligent data systems in modern space missions that increase efficiency, enhance mission autonomy and allow spacecrafts to make real-time decisions in dynamic and unpredictable environments where ground intervention can be limited. The first lecture will focus on the different obstacles that can be tackled with onboard machine learning algorithms and how satellite and ground systems can be modified to support them. We will then discuss the process of implementing such algorithms with brief code demos using an Arduino to simulate an onboard computer. The second part of the lecture will focus on the challenges faced in implementing such intelligent data systems in current space technology, the trade-offs amongst microprocessors used in existing spacecrafts for it, and the possible research required to develop feasible solutions for it.

Data in Motion: Inside Time-Series Databases

• Matthias Braeger (CERN)

From monitoring infrastructure at CERN to tracking financial markets, time-series data is everywhere—and growing fast. This talk takes you inside time-series databases: how they ingest massive write loads, optimize storage, and enable fast analytical queries over time. Expect practical insights, trade-offs, and a clear mental model for working with data in motion.

Large Language Model fundamentals

• Cristian Schuszter (CERN)

This lecture is aimed at giving participants a fundamental understanding on how statistical machine learning models have gotten really good (at an uncanny level) at generating quality text over a short timespan of just a few years. The talk will go over the fundamental building blocks, the major breakthroughs that were needed in order to reach our current level of model quality, and the new "agentic AI" paradigm that is taking root in software engineering as models start being able to interact with the outside world through a unified interface.

Technology Changes, Human Nature Doesn't (Part 1)

• Eveline Sintnicolaas (CERN)

As we race toward an era of generative AI and hyper-automation, we often mistake the speed of our tools for the speed of our evolution. While technology undergoes exponential shifts, the fundamental architecture of human behavior has remained largely unchanged for millennia. The mechanics of how we trust, collaborate, succeed are still governed by the same ancient social instincts and cognitive biases that kept our ancestors alive.

Vector Databases: The Long-Term Memory for AI

• Aryan Gupta

Today’s AI can answer questions, generate ideas, and assist in decision-making with remarkable fluency. But beneath this intelligence lies a fundamental limitation: memory. If databases are the brain of traditional systems, vector databases are the memory that makes AI truly intelligent. This presentation will take you on a journey into the memory layer of modern AI systems. We we discuss why do we need vector database in the first place. We will break down how vector databases enable machines to retrieve information based on meaning rather than keywords, allowing AI systems to remember and reason over vast amounts of data. Along the way, we will uncover how these systems power modern applications such as semantic search where relevant knowledge is retrieved and injected into AI responses in real time . By the end of this talk, you will see vector databases not just as another type of storage, but as the long-term memory that transforms AI from a static model into a dynamic system.

Making a Kubernetes Cluster, as the Heart of our Homelab

• Dennis Tsekouras (CERN)

What if you could self-host services like Google Drive, a password manager, or a DNS-level ad blocker on a cloud siting in your living room? In this talk, we’ll explore how to build a homelab powered by Kubernetes and turn old hardware into a resilient, scalable platform for running your own cloud-native applications. You’ll learn the core components of a Kubernetes cluster, practical deployment patterns, and how to securely access those services from the internet.

Technology Changes, Human Nature Doesn't (Part 2)

• Eveline Sintnicolaas (CERN)

As we race toward an era of generative AI and hyper-automation, we often mistake the speed of our tools for the speed of our evolution. While technology undergoes exponential shifts, the fundamental architecture of human behavior has remained largely unchanged for millennia. The mechanics of how we trust, collaborate, succeed are still governed by the same ancient social instincts and cognitive biases that kept our ancestors alive.

How RAG Enables AI Thinking in Real Time?

• Aryan Gupta

Today the advancement in Artificial Intelligence have brought us to a point where machines can assist in decision-making and interact with us in natural language. Yet beneath this impressive capability lies a fundamental limitation: these systems often rely on static knowledge, frozen at the time of their training. Unlike humans, they do not naturally consult documents, verify facts, or adapt to new information in real time. This presentation explores how modern AI systems overcome this limitation through a powerful architectural pattern known as Retrieval-Augmented Generation (RAG). We will take a journey inside the memory mechanism of AI systems, where external knowledge is no longer optional, but essential. Instead of relying solely on what a model has learned, RAG enables systems to actively retrieve relevant information from external sources before generating a response.

How we rebuild a legacy monolithic platform at CERN and still sleep at night

• Dmitry Kekelidze (CERN)

How do you modernize a critical legacy system inside one of the world’s most complex scientific institutions — while keeping your users, your stakeholders, and your sanity intact? At CERN, we’ve been rebuilding our enterprise document management platform — a core system supporting HR, finance, and everyday operations — moving from a large, tightly-coupled Java application to smaller backend services, a modern React web interface, and a workflow engine that coordinates business processes. But this talk isn’t about chasing shiny tech. It’s about making solid engineering decisions under real constraints: how to split a nearly 30-year-old system into manageable parts that teams can own; how to introduce clearer release and versioning rules so shared components don’t break each other; how to move from “everything talks directly to the database” to safer API-based integration; how to replace an embedded workflow component with a dedicated BPMN platform; and how to redesign a stateful, server-driven UI into a stateless React frontend backed by modular APIs. This isn’t a silver bullet — it’s a practical story of trade-offs, pitfalls, and patterns that can help you think about modernizing real systems under constraints, even in smaller projects.

Placeholder for Jan: Talk 2

• Jan Janke (CERN)

to-do

Your Brain is Lying to You: Fixing Bad Decisions with Analytical Thinking

• Sara Esperto (CERN)

Decision-making is often perceived as a rational and objec- tive process. However, extensive research in cognitive psychology demon- strates that humans also make decisions based on their judgment, which is systematically influenced by cognitive biases - recurring patterns of de- viation from rationality. These biases arise from mental shortcuts, which allow us to process information efficiently but can compromise accuracy. Even highly skilled professionals and the highest IQ scored people are therefore susceptible to flawed reasoning, leading to suboptimal or incon- sistent decisions. This paper argues that cultivating an awareness of these cognitive limitations and combining it with disciplined and evidence- based reasoning will consequently improve the way our brain regulates the quality of our decisions. This is particularly relevant in professional environments where analytical thinking plays a key role and will be the fuel for continuous improvement.

From Laptop to Data Center: Operating Databases at scale

• Andrzej Nowicki (CERN)

Modern software solutions rely on databases that are expected to operate continuously, serve geographically distributed users, and remain trustworthy under constant load. In such environments, a database is no longer a local component or a short‑lived project, but a long‑running service with explicit availability, performance, and reliability expectations. This lecture explores what fundamentally changes when databases move from development environments to always‑on services. We will examine real‑world constraints such as hardware failure, network instability, data growth, and human error, and how these factors shape operational practices. Topics include monitoring and observability, distinguishing symptoms from root causes, and understanding why many production problems cannot be detected through simple resource metrics alone. The session emphasizes practical lessons from operating databases at scale, showing why reliability must be designed in from the beginning.

Why Is My Query Slow? Database Indexes and the Query Planner Explained

• Batuhan Lel

Most developers learn SQL by writing queries. Very few learn why some of those queries are thousands of times slower than others — or how to fix them. This lecture fills that gap, giving students a working mental model of how database indexes are structured, how the query planner decides whether to use them, and how to read the evidence when something goes wrong.

Agentic AI Development Workflows

• Cristian Schuszter (CERN)

This lecture is aimed at being a more hands-on continuation of the LLM fundamentals course. Now that you are familiar with the building blocks of the large language model revolution, it is time to apply this knowledge to your own development workflows. We'll be covering MCP and ACP, two protocols that enable a wide array of use-cases, allowing models to interact with each-other and with specially defined APIs that are then used in order to generate more precise answers or run specialized tools. Lastly, we'll look into more advanced agentic orchestrations workflows, allowing you to effectively become a project manager of a team of agents that all work collaboratively or in a workflow to accomplish a more complex task.

Beyond Prompting: Agentic AI for Power BI Development.

• Christina Tsiplaki Spiliopoulou (CERN)

What if prompting was only the warm-up? AI is moving beyond chatbots. This talk explores how AI is evolving from prompt-based assistants into agents that can use tools, read context, and support real development tasks. Using Power BI development examples, we will look at what works, what fails, and what agentic development really looks like behind the hype.

Breaking (into) a Kubernetes Cluster

• Dennis Tsekouras (CERN)

The best way to understand Kubernetes security is to try to break it. Kubernetes is the core of modern cloud infrastructure, but common misconfigurations can quickly turn it into an malicious actor's playground. In this talk, we will go through the basic steps of a penetration testing procedure and explore different common misconfiguration and the damage they can provoke. Through a live demonstration of a deliberately misconfigured cluster, we will exploit these weaknesses step by step and show how seemingly small mistakes can lead to full cluster compromise, as well as best practices to prevent them.

Intelligent Data Systems and AI for Autonomous Space Technology (Part 2)

• Sudeshna Sengupta

Space technology has always been a fascinating topic that encompasses various fields of study. The current major space mission, Artemis II, to fly by the moon, has gripped the world’s attention. However, the amazing advancements in the field come with their own set of challenges now. Space has become congested and resource-constrained, which has now made the need for efficient and sustainable spacecraft operations more critical than ever. Most traditional space missions rely heavily on transmitting large volumes of raw data to Earth for processing, which requires extensive resources and is inefficient. But recent advances in onboard computing and artificial intelligence are reshaping the way such missions can process, filter and interpret data directly in orbit. Missions such as ESA’s ɸ-Sat series and OPS-SAT show how machine learning algorithms can identify relevant data in orbit, such as anomalous space weather and disruptions from space debris, which significantly reduces unnecessary data transmission. In doing so, these systems can lower power consumption, optimise bandwidth usage, and enable smaller satellites to perform complex tasks. This topic is a two-part lecture which explores the growing role of intelligent data systems in modern space missions that increase efficiency, enhance mission autonomy and allow spacecrafts to make real-time decisions in dynamic and unpredictable environments where ground intervention can be limited. The first lecture will focus on the different obstacles that can be tackled with onboard machine learning algorithms and how satellite and ground systems can be modified to support them. We will then discuss the process of implementing such algorithms with brief code demos using an Arduino to simulate an onboard computer. The second part of the lecture will focus on the challenges faced in implementing such intelligent data systems in current space technology, the trade-offs amongst microprocessors used in existing spacecrafts for it, and the possible research required to develop feasible solutions for it.

You Can't Debug Production: Logs, Metrics, and Traces Explained

• Batuhan Lel

When a service behaves unexpectedly in production, there is no debugger to attach, no breakpoint to set, and no console to inspect. The only window into what is happening is the telemetry the system was instrumented to emit. This lecture introduces students to the three pillars of observability — logs, metrics, and distributed traces — and gives them the tools to instrument a real system and interpret what it tells them.

Designing Robots for the Human Brain: UX in Autonomous Systems

• Sara Esperto

As autonomous systems become increasingly integrated into industrial, research, and everyday environments, the challenge is no longer only about building capable machines, but about designing systems that humans can understand, trust, and effectively collaborate and communicate with. Despite significant progress in autonomy and AI, many failures arise from a mismatch between machine behaviour and human cognitive expectations. This talk explores the role of Shared Mental Models (SMMs) in safe Human-Robot Interaction (HRI), and how UX designing principles and cognitive psychology can be combined to design functional autonomous systems that are also predictable and trustworthy in real-world contexts like CERN (e.g. Robodog, TIM).

Expecting the Unexpected: How do Backups & Recovery Work in Databases

• Andrzej Nowicki (CERN)

Failure is not an exception in large‑scale data systems. It is an expected and recurring condition. Disks fail, nodes crash, networks partition, and systems must recover correctly without losing or corrupting data. In scientific computing environments, recovery is not just about uptime, but about preserving correctness, reproducibility, and trust in results. This lecture focuses on how database recovery works in practice, using write‑ahead logging (WAL) as central concepts. We explain how WAL enables atomicity and durability, how it is generated and stored, and how databases distinguish between committed and uncommitted changes after a crash. The session walks through typical recovery scenarios, illustrating how logs are replayed. We will also cover how to run an on-line backup of a constantly changing database.

Placeholder: Lecture Given by Tartu University

• Veronika Zadin (University of Tartu (EE))

Examination

Prizegiving Ceremony and Closing Remarks

• Jan Janke (CERN)
• Matthias Braeger (CERN)