Perelyn

🎉 Proud Moment for Perelyn! We’re excited to share that a paper co-authored by our colleague, Dominik Filipiak, has been accepted to #AAAI 2025, one of the world’s top AI conferences organized by the Association for the Advancement of Artificial Intelligence (AAAI). Titled “A Wiener Process Perspective on Local Intrinsic Dimension Estimation Methods,” the paper explores state-of-the-art LID estimation methods and formalizes their behavior when assumptions fall short. It’s an important contribution to the field, and we’re delighted to see Dominik representing Perelyn alongside his talented collaborators: Piotr Tempczyk, Adam Kurpisz, and Lukasz Garncarek, PhD. This achievement reflects the collaborative efforts of the entire team and highlights Dominik’s valuable contributions to advancing AI research. We’re proud to have him as part of the Perelyn team, driving innovation in the field. Congratulations to Dominik and his collaborators on this outstanding milestone! 🚀 🔗You can find the pre-print of the paper linked in the comments below. #AIResearch #AAAI #LIDEstimation #GenerativeModels

📡 The Future of Telecommunications is Here: Powered by AI 🤖 The telecom industry is at a turning point. With 5G rollouts, skyrocketing mobile data usage, and the demand for gapless connectivity, the pressure to innovate has never been greater. And AI is the driving force making it possible. At Perelyn, we see how AI is revolutionizing the telecommunications industry: ✨ Smarter Networks: Predictive analytics ensure proactive maintenance, optimized bandwidth, and consistent performance during peak usage. 🌐 Better Customer Experiences: AI-powered chatbots, built on advanced Large Language Models, deliver personalized, efficient, and proactive customer support. 🛡️ Stronger Security: AI protects sensitive data by detecting and mitigating cyber threats in real time. 📊 Actionable Insights: From streamlining operations to crafting winning strategies, AI turns raw data into competitive advantages. Transformation isn’t without its challenges—integrating AI with legacy systems and ensuring data privacy requires expertise and strategy. That’s where Perelyn steps in. We work with telecom providers to implement impactful AI solutions tailored to their unique business needs. 🌟 The future of telecommunications is here. Explore Martin Griessmann's in-depth article on how AI is transforming the industry—find the link in the comments!

Earlier this year, at the prestigious [ICML] Int'l Conference on Machine Learning 2024, our contributions to the ICML 2024 Topological Deep Learning Challenge earned us 2nd and 3rd place in respective categories. Now, the results and approaches have been published in the paper "ICML Topological Deep Learning Challenge 2024: Beyond the Graph Domain." The paper delves into Topological Deep Learning (TDL), exploring how data can be represented across advanced topological domains like hypergraphs, simplicial complexes, and combinatorial complexes. Our hypergraph-based solution, designed to optimize information flow, demonstrates real-world applications in: 🟠 Medical Knowledge Graphs 🟠 Logistics Optimization 🟠 Complex Business Workflows If you’re curious about leveraging TDL for higher-order data relationships, this paper is a great resource. Huge congratulations again to the team: Maximilian Schattauer, Liliya Imasheva, Dominik Filipiak and Michael Banf for their remarkable work! 💡 Curious to learn more? Check out the link in the comments! #TopologicalDeepLearning #AIInnovation #MachineLearning

🎥 Catch the Recording: Maximilian Schattauer’s Session at the Prompt Engineering Conference Missed the chance to join Max Schattauer’s online session? The recording of his talk, "Improving Retrieval Q&A Contextualization Prompts," is now available! Max delves into the challenges of ensuring chatbot and Q&A systems retrieve accurate information by leveraging the entire conversation history, not just the latest user query. His session covers: ✅ Strategies for transforming message history into cohesive retrieval queries ✅ Using TextGrad to optimize query contextualization and prompt design ✅ Real-world examples showcasing how improved contextualization delivers smarter, industry-grade chatbot solutions This is a must-watch for anyone looking to enhance AI-driven retrieval systems and customer satisfaction. 👉 Watch the session now! The link to watch is available in the comments below. #PromptEngineering #TextGrad #AIInnovation #RAG

🔒 AI Security: Protect Your AI Investments in 2024 With AI becoming a critical part of business operations, security is no longer optional—it's essential. In fact, today 62% of organizations run AI packages with known vulnerabilities, leaving them open to unique threats that traditional IT security cannot fully address. From exposed API keys to overly permissive identities, unsecured AI models leave businesses open to potential data leaks, model manipulation, or even full-scale compromise of their AI infrastructure. Forward-thinking businesses have already started implementing robust AI security frameworks that will help them protect their data and be compliant with evolving regulations such as the EU AI Act. At Perelyn, we help organizations secure their AI systems from end to end. Our solutions in AI security minimize risks, secure your data, and make sure your business can innovate safely within the landscape that's moving so fast. Ready to protect your AI? Let's talk about how Perelyn can help you with this. For further discussion on the next steps to lock down your AI systems, please contact directly Johannes Kuhn, our AI security expert. Source: World Economic Forum, Orca Security #AI #AICompliance #AIsecurity #DataSecurity

We’re excited to share that our colleague, Martin Griessmann, recently appeared on the Rocket Beans Entertainment GmbH podcast Almost Daily, discussing Quantum Physics and AI. 🎙️ It’s always inspiring to see team members bring their expertise to broader audiences and engage in meaningful conversations about these fascinating fields. The podcast is a great watch for anyone curious about the intersection of AI and Physics, presented in an accessible and engaging way. This appearance comes shortly after Martin’s graduation from the University of Cambridge with a Master of Advanced Study in Physics—an achievement that underscores his deep commitment to pushing the boundaries of knowledge. At Perelyn, we value innovation and the skill to make complex ideas accessible—qualities Martin brings to the table every day. We’re proud to have him on our team and look forward to more exciting contributions! 📌 Link to the podcast in the comments. Please note: the podcast is currently available only in German. #TeamPerelyn #AIConsulting #QuantumPhysics #Innovation

We’re thrilled to announce that our very own Maximilian Schattauer & Johannes Kuhn will be speaking at this year's Prompt Engineering Conference on November 20th! 🎤 The session, "Improving Retrieval Q&A Contextualization Prompts", dives into one of the key challenges in building smarter chatbots and Q&A systems: ensuring retrieval queries include all the necessary context from conversation history. Why does this matter? Often, the context needed for accurate document retrieval isn’t just in the latest user question—it’s spread across several past interactions. By designing better prompts and leveraging tools like TextGrad, Max will share strategies to: ✅ Improve contextualisation in retrieval queries ✅ Avoid issues like missing context or irrelevant responses ✅ Deliver industry-grade solutions that boost customer satisfaction Expect real-world examples, actionable insights, and a showcase of how advanced prompt engineering transforms chatbot performance. Don’t miss this session! 👉 Details and link to the session can be found in the comments below. #PromptEngineering #TextGrad #AIInnovation #RAG

We’re excited that our Chief AI Scientist, Michael Banf, will be speaking at the Neo4j Developer Conference - #NODES2024 on November 7th! Michael will showcase how hierarchical graph structures can transform information retrieval, boosting precision and efficiency across industries. In his session, "More Patience for Patients: How LLMs and Graphs Enable Doctors to Focus on What They Signed Up For", Michael will discuss our work with Adiu health, where knowledge graphs help automate doctor-patient consultation documentation—freeing up time for essential patient care and research. We can’t wait to see how this innovation can impact not only healthcare but also other sectors. Join us to learn more and explore the potential of advanced knowledge graphs in driving industry advancements. #NODES2024 #GraphRAG #AIInnovation #Perelyn

🚀 The Age of LLMOps is Here! With approximately 6️⃣2️⃣% of machine learning teams now deploying or planning to deploy large language models, it is no surprise that 4️⃣2️⃣% are already adopting LLMOps tools in order to manage these powerful AI systems. But where does your company stand? Why LLMOps Can't Be Ignored With LLMs bringing transformation into industries, they also come with some complex challenges: scalability issues, security risks, and unpredictable costs. Without LLMOps, companies would run into unwanted, skyrocketing operational expenses, poor model performance, and the risk of non-compliance in heavily regulated sectors. Therefore, embracing LLMOps means monitoring, optimizing, and scaling your models. The days are gone when LLMOps was something nice to have; today, it is a must-have to keep your AI-driven operations running reliably and economically. What Are You Missing? 🟠 Security risks: The LLMs used to process sensitive data require maximum-security measures. LLMOps will provide you with the necessary governance of your models. 🟠 Model drift: If one is not actively monitoring model performance, it may degrade over time. The features of LLMOps include observability and automated fine-tuning ensure your AI stays sharp. 🟠 Operational bottlenecks: The deeper the models get, the more negligent their management might get-costly delays and missed opportunities. LLMOps smooths out processes so your AI operates at the pace of business. Perelyn's LLMOps Services At Perelyn, we offer bespoke LLMOps solutions that will have you scaling your LLMs with ease while minimizing risks and costs. From observability tool integrations to optimization for security and performance, we will make sure your AI systems are future-ready. Take the next step in your AI Operations with LLMOps. Have a conversation with us about how we can help you stay ahead of the game. You can visit our website or contact our LLMOps lead, Maximilian Schattauer, directly on LinkedIn. Source: "Large Language Model Adoption Reaches Tipping Point" Survey conducted by Arize AI #LLMOps #AI #MachineLearning #Perelyn #AIOptimization #FutureOfAI

🔍 Did you know? A staggering 97% of people globally support the principles behind trustworthy AI, and 71% believe that AI should be regulated. Customers are not just passively accepting AI—they're actively demanding that it be fair, transparent, and secure. Why Trustworthy AI is Now a Business Essential As AI becomes more integral to products and services, consumers and regulators alike are expecting companies to align with principles of trustworthy AI. Customers want AI systems that respect privacy, avoid bias, and operate transparently. Failing to meet these expectations can result in lost trust and missed opportunities. The Competitive Edge of Trustworthy AI Companies that prioritize trustworthy AI can differentiate themselves in the marketplace. It's not just about compliance—it's about building stronger relationships with customers who care about the ethical use of technology. Businesses that meet these expectations are more likely to gain customer loyalty and stand out in an increasingly crowded AI-driven landscape. How Perelyn Can Help At Perelyn, we understand the importance of delivering AI solutions that align with customer values and expectations. Our Trustworthy AI Services ensure your AI systems are built with fairness, transparency, and accountability from the ground up—giving you the competitive advantage that customers now expect. Are you ready to build AI that your customers can trust? Visit our website or connect with our lead, Johannes Kuhn, for more information. Source: KPMG "Trust in Artificial Intelligence" Study #TrustworthyAI #AICompliance #Perelyn #CustomerFirst #AIInnovation