Povijest Weblice

Every digital product follows a lifecycle - from idea and development to production and scaling. At each stage, legal risks emerge that tech teams often recognize too late. What does the EU AI Act mean for your AI-powered feature? Who is actually liable when your cloud provider suffers a breach? How do data protection rules shape your DevOps pipeline? And why are the NIS2 Directive and the Cyber Resilience Act - which applies to virtually every developer shipping digital products - no longer just a problem for "regulated industries"? Two lawyers walk you through three key regulatory layers - artificial intelligence, cloud computing, and data security & privacy - across the full stack, from your codebase to your cloud infrastructure. No legalese, real-world examples, and a practical checklist you can apply the moment you get back to your desk.

What does it really take to build a secure cloud-native SaaS platform for smart transport? In this talk, we’ll share our real-world experience designing and running a Kubernetes-based platform with security as a first-class concern. We’ll cover key architectural decisions, common pitfalls, and practical security challenges — from protecting services and data to maintaining resilience at scale. Expect honest lessons learned from production, what worked, what didn’t, and actionable insights for engineers building modern, security-focused SaaS platforms.

Last year, we explored how “vibe programming” enables developers to build applications using natural-language prompts instead of code. This year, we take it a step further: can those AI-generated applications actually be deployed to production?

In this session, I’ll share practical lessons from real-world projects where vibe-coded apps moved beyond prototypes and into live environments. Together we’ll look at:
• A live demo of taking an AI-generated app from prompt to deployment
• Proven workflows for refining, testing, and securing AI-written code
• Common pitfalls (and how to avoid them) when moving from demo to production
• How to blend AI speed with human oversight for reliability at scale

By the end of the session, you’ll see how vibe programming isn’t just for rapid prototyping—it can also deliver production-ready applications, faster than ever before.

UI asistenti/CoPilot-i pružaju koristi radnicima - automatiziraju repetitivne zadatke, izdvajaju uvide i povećavaju individualnu produktivnost - a istovremeno se suočavaju s "trikovima" koje UI izvodi, poput pristranosti, halucinacija i problema s povjerenjem.
Međutim, pojavila se ključna razlika u percepciji: zaposlenici intenzivno eksperimentiraju s umjetnom inteligencijom, dok mnogi lideri zaostaju u strateškom prihvaćanju umjetne inteligencije, podcjenjujući njezin potencijal i ne uspijevajući skalirati pilot projekte u značajne promjene unutar tvrtke.
Pridružite nam se u raspravi o tome što je potrebno da vodstvo od opreznog promatrača postane kompetentni praktičar – usklađujući primjenu umjetne inteligencije s poslovnim ciljevima, ulažući u razvoj vještina i uspostavljajući učinkovito upravljanje sve s ciljem omogućavanja tranzicije koja će premostiti taj jaz i u potpunosti osloboditi potencijal cijele organizacije!

Kako dizajnirati softver koji koristi tisuće ljudi s različitim razinama digitalne pismenosti, a s istim osnovnim zadacima - od rješavanja IT problema do upravljanja resursima?

Govorit ću o tome kako smo dizajnirali i razvili interni self-service portal. Cilj je bio omogućiti svakom od preko 12.000 zaposlenika globalne korporacije da samostalno rješava zahtjeve koji su se ranije čekali tjednima, mjesecima - ili nikad nisu bili riješeni.

Pokazat ću kako smo postigli ravnotežu između jednostavnosti korištenja za ne-tehničke korisnike i snage funkcionalnosti za napredne. Radilo se o složenom sustavu koji se oslanjao na više naslijeđenih alata i izvora podataka - često neurednih, nepouzdanih ili međusobno neusklađenih.

Ovo je priča o UX dizajnu u stvarnim uvjetima gdje ne postoji rješenje koje savršeno odgovara svima, ali se može postići ravnoteža koja pokriva ono što je većini zaista važno. Cilj je bio stvoriti sustav koji je pouzdan, jasan i dovoljno jednostavan da se koristi bez frustracije - bez nepotrebnih „žuljeva“.

Potrošnja AI tokena raste eksponencijalno. Agentski procesi, integracije konteksta i korištenje AI API-ja po radniku se umnožavaju stopama koje će današnje AI budžete učiniti zanemarivim za dvije ili tri godine. U međuvremenu, proizvodnja čipova, dostupnost memorije i GPU-ova skaliraju linearno u najboljem slučaju — a za izgradnju fabrika treba minimalno četiri godine, kapaciteti HBM-a (High Bandwidth Memory DDR5) su rezervisani od strane velikih igrača do 2028., a cijene memorije bi trebale porasti preko 50% do 2027.
Većina timova reaguje prebacivanjem na jeftinije modele ili skraćivanjem promptova. To je privremeno rješenje. Strukturalni odgovor je arhitekturalni: prestani obrađivati tokene koji ti uopće ne trebaju.

Ova prezentacija predstavlja višeslojnu, potpuno lokalnu RAG arhitekturu dizajniranu oko jednog principa — svaka komponenta u pipelineu postoji da smanji ono što LLM mora obraditi, dok poboljšava kvalitet onoga što zapravo vidi. Proći ćemo kroz produkcijski Data Warehouse sistem koji radi potpuno lokalno sa open-source modelima: od obogaćivanja podataka u vrijeme unosa tako da pretraga bolje funkcioniše kasnije, do slojevitih strategija pretrage koje hirurški biraju kontekst iz stotina mogućih izvora. Pokriti ćemo kako rastaviti kompleksna pitanja prije nego što pogode sloj pretrage, kako usmjeravati različite zadatke na različito dimenzionirane modele umjesto da se sve šalje na najveći.
Usput ćemo se osvrnuti i na druga ograničenje koje se krije iza problema troškova: privatnost. Za regulirane industrije pod GDPR-om i EU AI Aktom, slanje internih podataka na eksterne API-je nije opcija. Dobro izgrađen lokalni RAG sistem rješava oba problema istovremeno — manje tokena i potpuna suverenost nad podacima

Dobro izgrađen lokalni RAG sistem rješava oba problema istovremeno — smanjuje broj tokena koje LLM mora obraditi, i osigurava potpunu suverenost nad podacima. Umjesto da jurimo za jeftinijim modelima ili kraćim promptovima, pravi odgovor leži u arhitekturi koja eliminira nepotreban rad prije nego što uopće dođe do modela.

This presentation explores why Agentic AI is fundamentally restructuring software engineering. We begin by clarifying the "intelligence gap"—distinguishing between the reactive nature of LLMs and the autonomous, goal-oriented nature of Agentic systems.

Moving beyond theory, we’ll analyze how this paradigm is being operationalized within Big Tech. Is the shift toward "Agent-First" development a genuine evolution of the SDLC or just another high-cost hype cycle? We’ll examine the transition from writing code to Mechanism Engineering, where developers orchestrate fleets of agents to solve complex, multi-step problems. Whether you view it as a threat or a tool, the move from "AI-assisted" to "AI-autonomous" is no longer optional; we will conclude by determining if this shift will stand the test of time or buckle under the weight of enterprise reality.

How could AI affect solutions? Solutions have AI in them and are not "fixed" but adapt now! How should Solution Architecture for these new types of solutions look like? Also we are talking about architecture as code. Of course the next question is can agents “vibe” Solution Architecture? It is code now, no?
All these questions started from a LinkedIn article, so let’s go together on what it says, and see what we have now and what can wait us in the immediate to distant future. And some ideas on how to use AI to get us a nicer Solutions Architecture, from documentation to diagrams. Last but not least, how can Copilot help us create a Solution Architecture ? Yes it can also that...

Kako bi umjetna inteligencija mogla utjecati na rješenja? Rješenja sadrže umjetnu inteligenciju i nisu "fiksna", već se sada prilagođavaju po zadatku! Kako bi trebala izgledati Solution Architecture za ove nove vrste rješenja? Također, govorimo o arhitekturi kao kodu. Naravno, sljedeće pitanje je mogu li agenti "Vibe" Solution Architecture? To je sada kod, zar ne?
Sva ova pitanja započela su iz članka na LinkedInu, pa idemo zajedno na ono što piše i vidimo što imamo sada, a što nas može čekati u neposrednoj ili dalekoj budućnosti. I neke ideje o tome kako koristiti umjetnu inteligenciju kako bismo dobili ljepšu Solution Architecture, od dokumentacije do dijagrama. Na kraju, ali ne i najmanje važno, kako nam Copilot može pomoći u stvaranju Solution Architecture? Da, može cak i to....

How do you support the exponential growth of your product without the exponential growth of your teams?

Portfolio management, Wardley mapping, and Domain-Driven Design are effective ways to “do more with the same” in fast-growing organizations, but going by the book won’t guarantee the intended outcome.

Join this session to learn why you need 9 allies, as well as many other things, to introduce these ideas in fast-growing organizations.

#WardleyMapping #DomainDrivenDesign #PortfolioManagement #Strategy

Using C# / .NET Technologies and IoT to curate and automate video footage of garden hedgehogs.

An experimental journey through hardware device access, cloud storage, serverless functions, and a public website.

With an insight into problem-solving changes necessary to bring the code cross-platform to run on Linux on a Raspberry Pi.

An inspiring talk about how a completely unplanned event led to several fluid and experimental systems being developed, hacked together, and evolved in a unique way.

AI today writes code, generates images, holds conversations, and makes decisions but how much does it truly understand? In this keynote, Dr. Karlo Knežević offers a snapshot of the current state of AI: from frontier models and agents to capabilities that sounded like science fiction just yesterday.

But beyond the possibilities, we'll look at the other side of the coin: the fundamental limitations imposed by mathematics and theoretical computer science, the ethical questions growing louder by the day, and the thin line between what AI can automate and what remains inherently human. Finally, Karlo shares lessons from Sofascore: how AI was adopted across the entire company and what they learned along the way.

Metoda učenja u kojoj AI uči mehanizmom nagrade i kazne naziva se reinforcement learning, a u praksi postiže izvrsne rezultate u kompleksnim područjima primjene. Ova grana strojnog učenja ima cijeli niz prednosti, a na raspolaganju imamo više modernih framework-a za reinforcement learning. Upoznati ćemo algoritme poput Q-learning, Deep Q-networks i drugih, te vidjeti kako ih možemo grupirati. Područja korištenje reinforcement learning uključuju robotiku, AI u igrama, marketing, upravljanje resursima itd. Primjer korištenja je i AlphaGo, prvi programa koji je pobijedio profesionalnog igrača strateške društvene igre Go, a radi se o igri koja je teža (za AI) nego šah.

Što se događa kada se umjetna inteligencija spusti iz clouda i preseli izravno na umrežene uređaje? Nastaje AIoT – nova generacija pametnih sustava koji uče, analiziraju i donose odluke u stvarnom vremenu.
Na ovom predavanju ulazimo u svijet inteligentno povezanih uređaja koji pokreću pametne gradove, autonomne sustave i digitalnu transformaciju industrije. Kroz konkretne primjere, stvarne scenarije i tehnološke uvide pokazat ćemo kako AIoT briše granicu između softvera i fizičkog svijeta — i zašto je upravo sada pravi trenutak da budete dio te tehnološke evolucije.

We once believed atoms were indivisible — until we learned they weren’t. Unit testing has followed a similar path. For years, we’ve treated “unit” as something small and sacred — a single class, a method, a function. But in real systems, that tidy definition often doesn’t hold up.

This talk is about what happens when the textbook rules meet messy, real-world code. We’ll rethink what a “unit” really is, explore better ways to test meaningful behavior, and question some long-standing assumptions. Along the way, we’ll take a peek at how AI might help us write better tests — or at least ask better questions.

If you’ve ever wondered why your tests don’t quite match the way your code actually works, you’re not alone. Let’s talk about it.

We were told that technology would free us. Instead, we’ve entered a cycle of relentless acceleration where the 'future of work' arrives every six months. We are progressing faster than ever, yet we feel more lost. In this talk, we explore the paradox of our current moment: Why 'hard skills' like coding are becoming secondary to 'human skills' like empathy, resilience, and critical thinking.

Drawing from [mention your sources/experience], we will look at how to navigate the 'Excited/Scared' duality. We’ll discuss why the most important tool for the next decade isn't a new AI model—it’s our ability to remain human in an increasingly algorithmic world.

While Retrieval-Augmented Generation (RAG) is the gold standard for unstructured data (like PDFs, emails, and Word documents), it is fundamentally the wrong tool for querying structured tabular data (Databases, Data Warehouses, ERPs).

While RAG excels with unstructured text, a metadata-driven Text-to-SQL approach remains the optimal architecture for secure, real-time, and performant corporate data analytics.

Instead, an architecture where the LLM acts as a Semantic Translator (Text-to-SQL) remains the optimal architecture for secure, real-time, and performant corporate data analytics.

Demo included!

Discover how Agentic AI is transforming modern applications by combining autonomous reasoning with cloud-native scalability. In this session, we'll explore how Microsoft Azure enables the development of intelligent, goal-oriented agents using the powerful AI Foundry framework. You’ll learn how to orchestrate complex workflows with Semantic Kernel SDK, a flexible toolset for building AI agents that can plan, reason, and act. Whether you're building copilots, chatbots, or enterprise workflows, this session will ground you in the practical realities of Agentic AI on Azure.

The best UX is often invisible. Users rarely praise what works smoothly but they immediately notice when it fails.
Invisible UX lives in the small, structural decisions that prevent confusion before it appears: clear hierarchy, predictable flows, well-designed defaults, error prevention, consistent interaction patterns. These choices don’t draw attention to themselves. Instead, they remove hesitation, reduce cognitive load, and allow users to stay focused on their goal rather than the interface.
This talk examines how absence becomes a measurable signal of quality. Fewer support tickets. Shorter onboarding. Lower error rates. Higher task completion. When friction is designed out early, products feel “intuitive” - even though that intuition is carefully engineered.
Through practical examples, we will identify the invisible layers of interface design that shape user experience without demanding attention.

We have been hearing about artificial intelligence for a while now. The advantages of AI, new technologies, way to utilize them… But it usually starts going in a different direction: Can something go wrong with AI? A very interesting question that always veers into the scenario from a certain cheesy 1984. SF movie starring Arnold Schwarzenegger.

No, ChatGPT will not destroy the world because we made it fix our ugly React code, but there are many interesting scenarios where irresponsible use of AI can lead to unforeseen consequences. You may have already experienced some of them without even knowing.

This talk intends to educate the audience on a different way to talk about AI and its potential dangers. We will go over some lesser-known, but still very interesting AI scenarios and thought experiments and see what kind of future and (potential) dangers of AI they predict, how to recognize these dangers in our own products, and most importantly, how to prevent the worst from happening.