Grzegorz Czaja

Recently, I was informed that Dom Development S.A., a major residential developer in Poland, suffered a cyberattack resulting in the unauthorized extraction of data from its IT systems. According to the official notification, my personal data may have been affected. What is particularly concerning is the breadth and sensitivity of the data categories involved. Based on the communication sent to impacted individuals, the potentially exposed data appears to include not only standard identification and contact details, but also PESEL, identity document data, bank account numbers, transaction history, financial information related to crediting, civil status, and health-related data. This represents a combination of high-risk personal data and special categories under GDPR. While the company states that it has complied with its obligations under Articles 33 and 34 of the GDPR, the timeline between the incident and direct communication to affected individuals highlights how critical incident detection, escalation, and transparent communication are as part of mature data governance. Timeliness in this context is not a formality, it is a core risk-mitigation control. An additional concern is the sequence of communication. In my case, I became aware of the incident through media reports and public social media discussions before receiving a direct notification from the company. From a data protection and incident response perspective, learning about a potential breach affecting one’s own data via third-party channels undermines trust and reinforces the need for prompt, direct, and prioritized communication with affected data subjects. From a risk management perspective, the response also raises questions about whether the remedial measures offered are proportionate to the potential impact of the breach. Beyond formal notifications and general guidance, there appears to be limited evidence of enhanced support, concrete remediation, or accountability measures that would meaningfully reduce downstream risk for data subjects. In incidents involving identifiers with long-term consequences, compliance alone may not be sufficient to restore trust. I am sharing this not to sensationalize a single incident, but to underline a broader point: organizations that process large volumes of sensitive personal and financial data must treat cybersecurity and data protection as core operational risks, not auxiliary IT concerns. Incident response maturity, realistic threat modeling, and timely communication are essential components of that responsibility. We should continue to expect high standards of accountability and transparency when it comes to the protection of personal data. Repeated exposure to such incidents erodes trust, even when formal compliance requirements are met. It is my understanding this case has already been notified to Urząd Ochrony Danych Osobowych and the DPA as always will keep up their good work of oversight and control Mirosław Wróblewski

10

Hybrid vs. Remote Tech Roles: AI/ML Engineers vs. Support Specialists & Power BI Developers The tech industry is evolving rapidly—not just in terms of innovation, but also in how and where we work. Today’s professionals weigh not only skills and salaries but also flexibility, career growth, and global opportunities. Let’s explore the differences between two rising paths: AI/ML Engineers (hybrid roles) and Support Specialists or Power BI Developers (fully remote, B2B roles). 1. AI/ML Engineers (Hybrid Roles) AI/ML engineers often work in a hybrid setup, balancing onsite collaboration with remote flexibility. Experience: Exposure to cutting-edge projects in AI, machine learning, and automation. Benefits: Access to on-premise infrastructure, in-person collaboration for faster problem-solving, and strong career growth in AI-driven industries. Expectations: Frequent brainstorming sessions, working with large datasets, and aligning closely with R&D teams. 2. Support Specialists & Power BI Developers (Fully Remote, B2B) These roles thrive in fully remote, business-to-business models, giving professionals the freedom to work globally. Experience: Client-facing problem-solving, data visualization, and business intelligence insights. Benefits: Greater work-life balance, flexibility in location, and opportunities to collaborate with international clients. Expectations: High communication skills, self-management, and delivering solutions across time zones. The Big Picture: Flexibility & Global Reach Hybrid roles like AI/ML engineering offer deep innovation and collaborative synergy. Fully remote B2B roles such as Support Specialists or Power BI Developers spotlight the rise of a borderless workforce, where location is secondary to skill. 👉 Whether hybrid or remote, both paths reflect the future of work: flexible, global, and driven by expertise. For all our current open roles see: https://lnkd.in/dNi53vX #AI #MachineLearning #PowerBI #RemoteWork #HybridWork #FutureOfWork

6

#PBIMMM Power BI Monday's Morning Map. Episode #09 - Poland with H3 Some statistics 1️⃣ 380 counties This is just base for testing. I was wondering - is it possible to use H3 in #AzureMapvisual in PowerBI? There were so many failures, but now I did one step ahead. What is your favourite feature in AzureMap in PowerBI? #WrittenByHuman #PowerBIAzureMap #PBICoreVisuals #MicrosoftCoreVisuals #AzureMapVisual #mct

18

3 komentarze

🎉 First speakers announced for Sphere.it Connect: Data Horizons! Łukasz Pasisz & Wojciech Szymański Talk: Domain-Driven Data Architecture: the Data Mesh concept in practice You’ve probably heard the buzz around Data Mesh – and maybe even some skepticism. But there’s still plenty of ground to cover! In this talk, Łukasz and Wojciech will: ✅ Explore Data Mesh in practice ✅ Share platform design decisions ✅ Discuss scaling self-serve data ✅ Reflect on lessons learned from real-world implementations Don’t miss it! 🚀 Link below 👇 #DataHorizons #SphereITConnect #DataMesh #DataArchitecture

8

1 komentarz

DAISI: Data Assimilation with Inverse Sampling using Stochastic Interpolants Andrae, Larsson, Takao, Landelius, and Lindsten https://lnkd.in/d745_D-R This paper introduces DAISI, a novel data assimilation (DA) algorithm leveraging flow-based generative models to improve state estimation in complex, non-Gaussian systems. The core idea is to use a pre-trained, stationary generative prior (think normalizing flow) to represent the system's state distribution. Observations are then assimilated via a guidance-based conditional sampling scheme within the latent space of the flow. Crucially, DAISI incorporates forecast information through a clever "inverse sampling" step. Here's the technical breakdown: 1. Generative Prior: A normalizing flow, `p(z)`, is pre-trained on historical data to learn a representation of the system's state distribution. This prior remains fixed during the DA process. 2. Inverse Sampling: Given a forecast ensemble `x_f` (obtained from a numerical model), DAISI maps these forecast states into the latent space of the pre-trained flow using an inverse mapping `z_f = g(x_f)`, where `g` is the inverse of the flow's transformation. This provides initial conditions in the latent space for the subsequent conditional sampling. 3. Conditional Sampling: Using the latent forecast ensemble `z_f` as a starting point, DAISI performs conditional sampling to obtain an analysis ensemble `z_a` that is consistent with the observations `y`. This is achieved by minimizing a loss function that balances the likelihood of the latent states `z_a` under the prior `p(z)` and the likelihood of the observations `y` given the corresponding states `x_a = f(z_a)`, where `f` is the forward mapping of the flow. The paper uses a stochastic interpolant to guide the conditional sampling, ensuring a smooth transition between the forecast and the analysis. The key advantage of DAISI is that it avoids retraining or fine-tuning the generative prior at each assimilation step. This makes it computationally efficient and suitable for real-time applications. The inverse sampling step allows encoding model dynamics into the DA pipeline without requiring explicit knowledge of the model's equations within the flow architecture itself. --- Traditional DA methods often struggle with non-Gaussian distributions and high-dimensional systems, relying on approximations like the Ensemble Kalman Filter (EnKF) with heuristic tuning parameters. DAISI offers a more principled approach by learning a data-driven prior that can capture complex, non-Gaussian features of the system's state space. By decoupling the generative prior from the forecast model, DAISI provides a flexible framework for integrating machine learning with traditional numerical models. This is particularly relevant for applications where the underlying physics are complex and difficult to model accurately, such as weather forecasting, climate modeling, and subsurface flow.

26

🎥 Trust in AI Starts with Clean Data Is your GenAI project ready to scale—or stuck in PoC? In my video for PwC Poland’s Cloud & Digital Hub, I share why scaling GenAI takes more than tech. It requires quality data, strong governance, and cross-functional collaboration from day one. Without trust, GenAI won’t scale. With the right foundations, it will. 💬 What’s helping (or blocking) trust in AI in your organization? Let’s discuss. 👉 Watch the full episode on the PwC Poland Cloud & Digital Hub https://lnkd.in/dpHM79XT #GenAI #TrustedData #PwCPoland #DataGovernance #AITransformation #AIatScale

84

6 komentarzy

Anthropic or Philanthropic? Not long ago, my company rolled out access to Opus 4.6. Calling it impressive would be a massive understatement. We’re building a self-service customer portal, and I picked up a fairly standard access management task (RBAC). A few years back, I’d have estimated such a service in ~2 months for a team of 3 engineers. With Opus, I built it alone in a week. In parallel, I’m dealing with another issue that was offloaded to an external team for implementation. Tickets are created, sprint planning done, now I’m asked to schedule a deep dive. At this point, it would be objectively faster to just implement it myself with Opus than to scale development by engaging an offshore team. And it’s not just speed. Opus genuinely surprised me. Example: database migrations. I had two scripts (schema + data). Testing rollouts and rollbacks is usually painful: hard DB access, fragile data, slow refreshes. Many teams I’ve seen simply don’t test rollbacks because the overhead isn’t worth it. Opus ran a full scenario for me using local docker: snapshot initial DB state, seed tricky test data, applied schema migration, data migration, rollback in reverse order, verifying each step. It produced a detailed report with table dumps, migration state, and metadata for each step. Frankly, that’s more than I’d ever expect from a good QA. I’ve written before about how AI changes the review calculus. The same applies to the engineering calculus. It now makes sense to schedule only big tasks and give them to engineers confident enough to deliver end-to-end. The real question becomes: Is it faster to implement than to involve someone else? This also puts classic Scrum ceremonies under extra scrutiny. As cycle times collapse, the overhead becomes harder to justify economically. If you’re an old-school engineer, sceptical of AI - this might be the right moment to revisit your views. If you havr a pet project or idea you could never quite prioritize or staff - dust it off and try Opus. Many here come from electronic trading and know Chronicle / Peter Lawrey. Read his recent posts: one of the strongest engineers in the field is openly pursuing agents development. You’re no longer resource-constrained. Feed your idea to Opus, and if you can validate and polish the output using your expertise, you can have it 90-95% done almost instantly. And finally: the economics are wild. At current pricing, using Opus feels almost unfair. Honestly, Anthropic could rename itself Philanthropic.

2

When we think of data applications, it’s easy to focus only on functionality, but what if we could merge analytical power with design that truly engages users? Learning to build Shiny apps with HTML, CSS, and JavaScript isn’t just about adding polish; it’s about reimagining how data stories are told and experienced. This course is designed to spark that shift in perspective: to see Shiny not as a static reporting tool, but as a canvas where interactivity, aesthetics, and insight come together to create something that resonates. Check comments for the link

2

1 komentarz

The dark side of Domain-Driven Design: When architectural purity becomes a liability Imagine this scenario: A team starts with a simple user registration feature. Six months later, they have 37 classes to create a user account. Six DTOs for name, email, and password. A domain event system that publishes events nobody consumes. They followed every DDD pattern in the book. They created aggregates for entities with no invariants. They built repositories with specification patterns for simple CRUD. Event sourcing for data that never changed. The result? A codebase so complex that adding a new field requires touching 12 files. Unit tests that take 3 minutes to start. New developers who spend their first week just trying to understand where business logic lives. This is what happens when DDD is followed dogmatically. --- THE PRAGMATIC APPROACH I've written a guide on doing DDD without the bloat, drawing from patterns I've observed across the industry. The core principle: Architecture serves the code, not the other way around. Here's the decision framework: ✅ Use DDD when: - Complex business rules with multiple workflow paths - Long-term projects (2+ years) with evolving requirements   - Multiple teams contributing to the same codebase ❌ Don't use DDD when: - Simple CRUD with no complex rules - Rapid prototypes or MVPs - Team is unfamiliar with the patterns --- START SIMPLE, EVOLVE AS NEEDED Don't architect for year 3 on day 1: Phase 1: Transaction Script (Weeks 1-4) Phase 2: Anemic Domain Model (Months 2-3) Phase 3: Rich Domain Model (Months 4-6) Phase 4: Full DDD (Year 1+, only if needed) Your domain layer should have zero dependencies on frameworks. No Spring. No JPA. No HTTP. Just pure business logic. Dependencies flow inward: Presentation → Application → Domain           ↑ Infrastructure ─────┘ --- KNOW WHEN TO SIMPLIFY The best engineers aren't the ones who apply every pattern. They're the ones who know which patterns to skip. Red flags of over-engineering: - Aggregates protecting nothing but trivial state changes - Use cases that do nothing but delegate to repositories - "Simple" changes require touching 5+ files I've compiled everything into a comprehensive guide covering migration strategies, testing approaches, common pitfalls, and practical Java 25 examples. The goal isn't architectural purity. It's building software that bends without breaking. "Good architecture allows you to change your mind without changing the world." Stop building cathedrals for vanity. Build houses people can actually live in. 🔗 Read the full guide: https://lnkd.in/dZndu-Ax

2

In recent years, IT investments faced a downturn due to the post-pandemic decline in demand. However, 2025 marks a significant turnaround, particularly within Poland's insurance sector. 🧠 According to the latest report from the Polish Chamber of Insurance (PIU), 80% of Polish insurers are increasing 📈 their budgets for IT and technological developments. The emergence of technologies like AI is propelling the market to optimize processes and address new challenges. Key areas of development include: - AI and Machine Learning: Personalizing offers, optimizing decision-making processes, and enhancing risk analysis. - Online and Mobile Portals: Facilitating direct insurance interactions. - Automation and Optimization: Streamlining policy administration and claims handling. - Chatbots and Voice Bots: Providing automated responses to customer queries. Notably, nearly half of the insurers surveyed have already integrated AI into some of their operations. Poland's insurance market is renowned for its adoption of modern technologies. While face-to-face 🤝 interactions remain prevalent in initiating insurance sales, clients frequently utilize online platforms like Rankomat to compare insurance prices and conditions. Insurers in other markets should take note. As younger generations enter the insurance landscape, the demand for modern, digital solutions is set to rise rapidly. At Sollers Consulting we've garnered extensive experience from the Polish market and are eager to assist you on your journey toward an efficient, digital future for your business. #CEEInnovation #InsuranceTechnology #DigitalTransformation #FutureOfWork

9

1 komentarz

Still using Oracle Forms? It’s high time to move away from legacy technologies. At Pretius, we’ve completed numerous Oracle Forms to APEX migrations over the past few years — and we’ve achieved a 90% estimation accuracy rate. 🚀 Want to benefit from our experience? Michał Lipiński introduces our custom AI-powered Forms to APEX Assistant, designed to support your migration journey: ✅ Accurately estimate your migration project using AI
 ✅ Kickstart the process with semi-automated conversion
 ✅ Get AI support for developers throughout the entire project 🎥 Check it out now on YouTube! #oraclapex #oracleapex #formsmigration https://lnkd.in/dExb77Mp

9

At Digmatix Poland, we often hear doubts about low-code platforms like Microsoft Power Platform. Let's bust the most common myths: ❌ Myth 1: Low-code isn’t enterprise-ready. ✅ Reality: Power Platform supports enterprise-grade apps that scale across entire organizations — not just small teams. ❌ Myth 2: Low-code is only for simple apps. ✅ Reality: Businesses build mission-critical solutions on low-code that integrate seamlessly with core systems. ❌ Myth 3: Professional developers won’t use low-code. ✅ Reality: Developers use Power Platform to speed up routine work, leaving more time for complex challenges — reducing development time by up to 50%. ❌ Myth 4: Low-code isn’t secure. ✅ Reality: Microsoft Power Platform meets the highest security and compliance standards, protecting enterprise applications just like traditional development. 🔍 Why does this matter? ✅ 224% ROI potential ✅ Up to 85% development time savings ✅ 60% higher success rates with Copilot in Power Apps Don't let outdated beliefs limit your digital growth. 📅 Let’s explore how Power Platform can empower your organization: https://lnkd.in/d2dqYWcB

9

Absolute #must #hear pov from Paweł Wrona from PwC #Data #Mesh Team - if you run or plan #Data #Transformation, focus on scaling AI or consider building a Data Mesh foundation for better business use of data! Data availability and data quality are critical for any #Analytics and #AI work - therefore it is fundamental, so we design and implement #Data #Technology in line with the #Business #Value it create and support. One more important often forgotten ingredient is the (good, old and so much under-valued) #Data #Governance. Data quality is as good as we manage it so when investing in new Data Platforms and AI-tools worth to plan also some effort (and budget) to solve the Data Catalogue & MDM in parallel. In PwC we address Data Transformation in a holistic manner - from Strategy & Architecture, through Data Technology Platform & Data Engineering, to Data Governance & Master Data Management. We work like one big Data Mesh Incubator so you can focus on your Business ideas! Contact us so we can help in your Data journey! Marcin Makusak Mariusz Walkiewicz Janusz Bogacinski Natalia Roślik Mariusz Chudy Paweł Kaczmarek ☁Michael Norejko Kamil Marcin Kosiński, PhD Marek Mlyniec Jakub Borowiec

21

In this year’s edition of the “Finances of Polish Women and Men” report, we took a closer look at the role of artificial intelligence: the kind we notice, the kind we sense intuitively, and the kind we don’t yet have a name for. Increasingly, people are aware of AI’s growing influence on their daily lives. However, they’re not always sure how to interpret it. This isn’t a technology that inspires only excitement. It’s a technology that demands reflection. And that’s exactly the point, as shown clearly in the chart I’ve included. This part of the report reveals a shift. We’re beginning to see change more clearly, and we’re no longer afraid of it. Still, there’s a shared feeling that, as individuals, we’re at a disadvantage and that only the largest players will benefit from this revolution. This doesn't have to be true if we begin to actively use and better understand AI. As an industry and as Autopay, of course, I believe we have a critical role to play. Not just in building tools, but in helping people understand them. Our reports aren’t just data sets. They are maps of behavior, emotion, and unmet needs. Helping us design solutions that are not only advanced but also meaningful. #autopayreports #AIimpact https://lnkd.in/d7SDrJQP

35

The rapid rise of open‑source LLMs like Moonshot’s Kimi K2 and K2.5 marks a major shift in enterprise AI. These models now deliver frontier‑level multimodal and agentic capabilities, powered by trillion‑parameter Mixture‑of‑Experts architectures and massive training corpora. But with this leap forward comes a new layer of risk: -Unpredictable model behaviour that can create legal, operational, and reputational vulnerabilities in production environments -Safety degradation through fine‑tuning, increasing exposure to prompt injection, data leakage, and adversarial misuse -Expanded attack surfaces, from poisoned training data to insecure output handling and shadow AI deployments. As organizations accelerate adoption, the question becomes: How do we unlock the value of open models without compromising trust, security or performance? At EY, our #QualityEngineering and #Cybersecurity teams are partnering to help clients safely scale generative AI—through rigorous model validation and red‑teaming, secure deployment patterns, responsible AI controls, and governance frameworks purpose‑built for LLM risk. Open‑source LLMs create enormous opportunity. With the right guardrails, they also create enterprise‑grade trust. Read our article: https://lnkd.in/dn6Wx2Y2 for more information. If you’re exploring how to deploy or govern these models safely, let’s connect. #AI #QualityEngineering #ResponsibleAI #SoftwareTesting #EthicalAI #TrustInTech #DevOps #DataIntegrity #ExplainableAI #Compliance #TestingStrategy #EY_Technology

43

8 komentarzy

Poland's Insurance Market: Local Champion vs. Foreign Giants Fascinating dynamics in Poland's insurance landscape! 7 out of 8 top insurers are foreign-owned, while market concentration is high and growing. Nevertheless, the local incumbent PZU continues to dominate with a solid 34% market share and PLN 15.1 billion in total gross written premiums (as of H1 2025, combined premium written of life & non-life subsidiaries).   - PZU's defensive moat - The Polish champion maintains its leadership position despite intense foreign competition, showcasing the power of local market knowledge and established distribution networks - Foreign powerhouses gaining ground - Talanx (16.6%), ERGO Group AG (12.1%), and Vienna Insurance Group (VIG) (9.7%) represent the European insurance giants successfully penetrating the Polish market. They are systematically trying to erode PZU's market dominance, evidenced by recent shifts such as Warta's displacement of PZU in the motor TPL segment leadership in early 2025. - High market concentration - The top 3 players control nearly 63% of total premiums, indicating a mature, consolidated market with significant barriers to entry - Strategic implications - This concentration suggests limited room for new entrants, while existing players focus on defending market share through innovation and customer retention - The Polish insurance market exemplifies how local incumbents can maintain leadership even in highly competitive, foreign-dominated sectors. PZU's resilience against well-capitalized European competitors demonstrates solid defensive strategy... but can it last as foreign challengers raise the stakes and heavily invest in automation, distribution, and other? #Insurance #Poland #MarketAnalysis #PZU #EuropeanInsurance #MarketConcentration #InvestmentResearch Want to learn more about the insurance market in Poland? Check the report: https://lnkd.in/d7HinKNj

4

📊 As Ivelina Terzieva says, environment-aware execution is key to reliable data workflows.

 ⚙️ Databricks Delta Live Tables supports rapid development and resilient production with built-in automation and recovery.

 🔗 Explore the best practices: https://lnkd.in/dNTqqyYE #Databricks #DeltaLiveTables #DataEngineering #ETL #DataPipelines #Automation #DataOps #DataQuality #Lakehouse #BigData #beye

13