Marc Plogas
Herding AI Cats, Before It Was Cool.
Berlin, Germany
Actions
Marc Plogas discovered his passion for technology at the age of six when an Atari sparked his interest in programming. He holds a Master's in Computer Science and spent nearly a decade freelancing as a software, mobile, and test engineer, as well as a solution architect, before joining Microsoft. At Microsoft, he played an instrumental role in pioneering technologies such as Windows Mixed Reality, Desktop Bridge, and WinUI 3, and later advised startups worldwide on optimizing their cloud architectures with a focus on reliability, scalability, cost efficiency, and security. Now at RoboTwin, Marc builds reliable cloud and edge IoT backbones for no‑code robot teaching and continues to deliver talks and workshops on IoT, AI, and cloud architecture.
Area of Expertise
Topics
Teaching Robots at Scale: Data, Programming, and the Real Limits of Industrial AI
Despite decades of investment in automation, many high-value industrial operations in manufacturing remain largely manual. The limiting factor is no longer hardware – lack of machines or robots – but software. Programming automation of complex tasks is slow, expensive, and dependent on expert and domain specific knowledge.
While AI has been transforming software systems towards higher autonomy, making production environments autonomous is fundamentally harder. Conversational AI can learn through access to massive datasets on the internet, however, there is no such treasure trove or robot data.
This talk will analyze why today’s AI successes in robotics focus on controlled, low-value tasks, and why expectations around humanoids and general-purpose robots remain misaligned with industrial reality. The core challenge is data generation: without systematic ways to capture manufacturing and production data, learning-based robotics cannot scale. And without measures to protect the data and control the learning process, the solutions cannot be trustworthy, nor safe. Our approach to reliable data foundation required for the adoption of AI in robotics will be shared with the audience to open discussions about its benefits, challenges, and opportunities.
The Agent Smith Problem: When AI Agents Turn Rogue
"The Best Thing About Being Me... There Are So MANY Me's!" is a famous quote from Agent Smith. With the advancements of mulit-agentic systems, we're witnessing agents that can spawn, coordinate, and replicate across enterprise environments - and just like Agent Smith, real AI agents carry can carry multiple security risks.
This session explores the emerging threat landscape where trusted AI assistants can be compromised, manipulated, or even weaponized; turning helpful tools into potential attack vectors that can escalate privileges, exfiltrate data, and propagate vulnerabilities across interconnected systems.
Herding AI Cats: Semantic Kernel Multi-Agents Scenarios.
Semantic Kernel: Beyond the "Hello, World."
This deep dive gets elbows-deep into the framework for serious LLM, plugin, and agent orchestration. We'll tear down a multi-agent software dev scenario to expose advanced techniques in orchestration, planning, and integration. Expect real code, real problems, and how to beat Semantic Kernel into submission for complex tasks. If you want the advanced stuff, this is it.
Semantic Kernel Agents: Who Needs Developers Anyway?
Semantic Kernel is a framework designed to facilitate the integration and orchestration of large language models, plugins and agents in AI applications. In this session, we start with a brief overview of Semantic Kernel core concepts and plugins followed with an in-depth description of Semantic Kernel Agents. To illustrate the concept of multi-agent scenarios and benefits of multiple AI models, we'll present a collaborative demo to streamline and enhance the software development process.
From RAG to Riches: The Evolution of Retrieval-Augmented Generation
Traditional RAG approaches have transformed the way we integrate content retrieval with generative models. However, they come with notable limitations such as efficiency bottlenecks, scalability issues, and challenges in context integration. In this session, we will explore these shortcomings in detail and introduce you to cutting-edge innovations like RAG 2.0 and GraphRAG. Learn how RAG 2.0 enhances retrieval mechanisms to improve performance and accuracy, and discover how GraphRAG leverages graph structures for superior context management and richer information synthesis.
Let's delve into the fascinating world of RAG and its latest advancements together!
Constructing a Semantic Ingestion Pipeline
Build a seamless, continuous ingestion pipeline capable of processing diverse data formats such as PDFs, images, and markdown using Semantic Kernels Kernel Memory service. This session will guide you through the intricacies of creating a system that not only scans and updates files at regular intervals but also optimizes chat applications with robust, contextually aware responses using Retrieval-Augmented Generation (RAG).
How to Train Your Prompt Dragon and Protect Your Model
As the power and utility of Language Model (LLM) technology continue to grow, so too does the importance of Prompt Engineering - or Prompt Programming - in guiding its application. In this session, we'll provide a quick recap on how LLMs work before diving into the world of Prompt Engineering. We'll examine different types of prompt engineering and weigh the pros and cons of their use. But with great power comes great responsibility, and the risks of "prompt demons" in the form of security concerns with prompt injection cannot be ignored. To address these issues, we'll discuss the best practices for secure prompt engineering, and explore the future of prompt engineering and security with LLMs. Join us for an informative session on navigating the balance between innovation and security with Transformer models.
GTP and Codex: The Dynamic Duo of NLP or Skynet's First Step?
Large Language Models (LLMs) have transformed the way we process natural language. OpenAI's GPT and Codex models have taken NLP to a new level, making it possible to perform tasks such as language translation, question-answering, and text generation with unprecedented accuracy.
In this talk, we will discuss the inner workings of Transformer models to help us understand their societal impact, including their impact on information retrieval, job markets, privacy, and the environment. We will also discuss the ethical concerns surrounding LLMs, such as potential biases and their impact on society at large.
Agent Smith Gets Hardware: Autonomous IoT Hacking From Debug Port to Cloud API
Last year, Agent Smith turned rogue in software. This year, he got physical access.
In this session, we take a cheap off-the-shelf IoT device and let AI agents loose on it. No manual hacking, no memorized commands, no Kali cheat sheets. Using MCP-driven tool orchestration, the agents autonomously probe the device's hardware debug interface, extract firmware secrets, and intercept its cloud traffic through a rogue WiFi access point. Multiple attack vectors, one device, zero human intervention.
This is not a theoretical exercise. Every demo runs on a Raspberry Pi with open-source tools, and the agents make their own decisions about what to probe, what to extract, and what to flag. We will see what they finds, what they miss, and what happens when a device actually fights back.
Whether you build IoT products, secure enterprise networks, or just want to know what your smart devices are doing behind your back: this session will make you uncomfortable.
Good.
Technical Summit 2026 Sessionize Event
AI Show and Tell - Berlin Sessionize Event
AI Community Day Cologne Sessionize Event
AgentCon 2025 - Berlin Sessionize Event
Global AI Community Bootcamp {Berlin} Sessionize Event
AI Community Day - Berlin Sessionize Event
Technical Summit 2024 EN Sessionize Event
Global AI Bootcamp 2024 Germany/Karlsruhe Sessionize Event
Microsoft Build 2023 After Party - AI & Metaverse Playground Berlin Sessionize Event
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top