Overview
ePotentia is a scientific and industrial AI consultancy based in Antwerp, founded in 2019. The legal entity is ePotentia VOF, registered at Frans Van Dijckstraat 59 in Antwerp. The company has 2-10 employees. The company started in web hosting and development, then launched the ePotentia concept around ecommerce resources and expanded into mobile app development. It later shifted to consulting focused on data science and machine learning. Today ePotentia builds AI workflows that combine models with expert knowledge, supported by explainable AI and uncertainty estimation. Services cover data collection, storage, analysis, validation and AI-ready processing, and models ranging from statistical models to deep neural networks with built-in uncertainty quantification. Results are visualised with SHAP values and uncertainty bands. Target applications are materials science and life sciences. The company also gives training workshops on statistics, machine learning and deep learning. Michael Sluydts, PhD, is Chief Executive Officer, with a background in deep learning and IT and in engineering disciplines related to materials and chemistry. Vincent Sluydts, PhD, is Chief Data Officer, with a statistical background and expertise in biological sciences and clinical trials.
Stated facts & numbers
- Founded: 2019
- Headquarters: Antwerp
- Legal entity: ePotentia VOF
- Employees: 2-10
- CEO: Michael Sluydts
- Chief Data Officer: Vincent Sluydts
- Focus: AI and data science consulting for materials science and life sciences
Key people
In the news
- OpenAI's Blueprint for Detecting AI Misalignment As AI models become more capable, detecting when their behavior departs from intended actions is critical. OpenAI has unveiled its multi-layered monitoring system designed to catch frontier model misalignment during training, evaluation, and deployment. Layered Defense and Real-World Impact The system uses a tiered architecture ranging from lightweight activation probes to model-based reviewers and investigative agents. This approach has already proven effective. During a recent
- Spooky Cyber Trends for 2026: AI as a Threat Multiplier As we approach Halloween, the cybersecurity statistics for 2026 paint a chilling picture. According to the CrowdStrike 2026 Global Threat Report, attacks by AI-enabled adversaries have surged by 89%, with the fastest breakout time dropping to just 27 seconds. AI is fundamentally changing the threat landscape, not necessarily by inventing new attacks, but by dramatically accelerating the speed and scale of existing ones. The Persistence of Basic Vulnerabilities Despite
- Google DeepMind Unveils Genie 4 World Model Google DeepMind has launched Genie 4 in research preview. The latest world model generates photorealistic, interactive environments complete with real-time audio. A major upgrade allows users to type prompts mid-simulation to change weather, adjust camera angles, and add objects on the fly for up to two minutes. Integration and Applications Genie 4 is tightly integrated with Gemini to plan and execute tasks, opening new doors for testing robotics scenarios in simulated environments.
- The Messy Origins of AI Hallucinations Generative AI models are rooted in the ideas of mid-20th century engineers who believed a machine needed a touch of irrationality to think like a human. A new paper by MIT assistant professor Benjamin Lindquist traces the history of AI hallucinations back to these early pioneers. Creativity Through Randomness To mimic the unexpected leaps of human creativity, early engineers intentionally injected randomness into their algorithms. Unlike today's developers who want predictable and correct
- Independent Evaluators Test OpenAI's Monitoring Systems Nonprofit evaluators METR and Redwood Research recently concluded a pilot review of OpenAI's internal monitoring systems. Granted "employee-like access" to OpenAI's codebase and documentation, the researchers were able to rigorously stress-test the company's safeguards. Identifying and Patching Vulnerabilities During the three-week pilot, the evaluators discovered several novel vulnerabilities. This included an exploit that would allow AI agents to prevent hundreds of recent
- AI Defends the Web: Discovering "php2shell" A 14-year-old software flaw threatening up to 70% of the internet has finally been uncovered. The vulnerability, dubbed "php2shell," was independently discovered by security teams at Calif, OpenAI, and Wiz using frontier AI models like GPT-5.6 Sol and Gemini 4 Argon. An Unprecedented Collaboration With millions of websites potentially exposed, OpenAI led a coalition of maintainers, researchers, and security providers, including the PHP Foundation and Cloudflare, to develop emergency
- MIT's "AlphaFold for Materials" Maps 3D Crystals MIT researchers have developed a groundbreaking multimodal AI model that combines 3D atomic structures with natural language descriptions. Traditional language models struggle with materials science because they rely on text, whereas materials exist as highly ordered 3D crystal arrangements. By training the AI using contrastive learning to "see" crystalline geometry and bonding patterns, the model successfully identifies optimal materials for specific applications, outperforming
- The Dawn of AI IoT Worms Anthropic has issued a stark warning: advances in small AI models could soon usher in a new era of self-replicating malware. In recent experiments, researchers demonstrated how small models could be distributed across low-cost Internet of Things (IoT) devices, such as Android TV boxes, to establish inference and propagate autonomously. The Autonomous Malware Risk Angle Unlike traditional malware written for specific hardware, these AI worms use their model weights as the payload, dynamically
Related profiles
Something wrong or missing? Send an update. Fixed within 24 hours.