Aignostics, a spin-off from Charité – Universitätsmedizin Berlin, develops AI-based solutions for pathology. The company relies on its proprietary “Explainable AI” platform, which is based on research conducted by the Fraunhofer Society, Charité and the Technical University of Berlin, to overcome the often-criticised “black box” problem associated with AI. Aignostics focuses on developing complex applications for pharmaceutical research and precision medicine.

x-cardiac uses deep learning methods (recurrent neural networks) to predict several serious complications (mortality, renal failure requiring renal replacement therapy, and post-operative bleeding leading to surgical revision) in post-cardiac surgery care in real time. Software analysis methods and artificial intelligence techniques are used to provide intensive care specialists with better information about patients at risk of complications. This makes it possible, for example, to identify at an early stage which patients are at high risk of serious complications following heart surgery.

The BigEarthNet dataset is a large-scale, multimodal remote sensing dataset created to support AI research in Earth observation, comprising 549,488 pairs of Sentinel-1 and Sentinel-2 satellite images. In addition, software tools have been developed to convert the dataset into a format optimised for deep learning.  BigEarthNet and the associated software tools have opened up promising avenues for advancing research into AI for Earth observation. Users include developers of Earth observation applications and services, space agencies, the space industry and the scientific community.

AI models are only as good as the data they act on — yet most IoT data never reach the cloud in time to matter. NebulaStream is an open-source, high-performance stream processing engine that places AI inference directly at the sensor, edge device, and gateway, unifying the entire sensor-edge-cloud continuum in a single programmable system. This reduces the time to action in critical situations and enables large deployments, as data is only sent on demand and processed by any device, not only in the cloud.  It supports diverse hardware architectures, compiles queries into hardware-optimized code, and enables continuous, multi-modal ML inference at scale. Initial applications, such as a prospective study in smart ICUs, smart transportation at SNCB and in the context of smart production are currently beeing implemented. NebulaStream bridges the gap between frontier AI research and real-world impact — with a growing open-source community and an Apache 2.0 license. NebulaStream already has earned several prestigious scientific awards, including Best Demo Honorable Mentions at ACM SIGMOD and a Best Paper Award at IEEE ICDE. We are currently looking to grow the NebulaStream ecosystem and explore tech transfer via companies and startups.

FAIRe (Frugal Artificial Intelligence in Resource-limited Environments) is a joint DFKI-Inria research project developing the scientific foundations of resource-efficient AI for edge devices. The project investigates methods across the full technology stack, from AI models and compiler technologies to specialized hardware. By combining expertise in machine learning, systems software and hardware, FAIRe enables advanced AI on devices with limited computing power, memory and energy resources.

HumAIne develops the next generation of human-centered AI, enabling reliable collaboration between humans and AI in dynamic real-world environments. Combining expertise in machine learning, explainable AI, neuro-symbolic reasoning, swarm learning, semantics and data analytics, the project creates hybrid decision-support systems. Its HumAIne Operating System integrates these technologies to help developers build transparent, trustworthy and user-friendly AI solutions that outperform either humans or AI working alone.

SOOFI develops open European foundation models, a sovereign alternative to non-European models, funded by the German Federal Ministry for Economic Affairs and Energy under the IPCEI-CIS initiative. The focus is on industrial applications where control, transparency, and efficiency are critical. Within SOOFI, Lamarr researchers contributed substantially to pre-training. The first model, Soofi S, is a Mixture-of-Experts hybrid Mamba Transformer trained on roughly 27 trillion tokens. SOOFI thus exemplifies how Lamarr's foundational research translates directly into industrially usable AI infrastructure.

SLAPStack brings artificial intelligence into everyday warehouse operations to eliminate inefficiencies, turning manual logistics into hybrid, automation-ready operations. Its computer vision camera-based system, TrackRack, retrofits onto any existing forklift, turning every vehicle into a real-time data source, capturing movement, inventory, and load dimensions without added site infrastructure. This continuous stream feeds SLAPStack OS, an AI-driven warehouse operating system that uses simulation and machine learning to optimize routing, storage allocation, and layout planning, and to power a growing, self-improving dataset. Designed as a modular SaaS solution compatible with any fleet, SLAPStack, the intelligence layer, bridges the gap between traditional warehouse operations and fully autonomous, self-optimizing logistics.

Physics-LLM develops an LLM-enhanced toolkit for research data management, enabling seamless publication and curation of scientific data while improving their findability and the effectiveness of scientific workflows — advancing physics research from Big Data to "Smart Data" under FAIR data principles. The project is funded within the BMFTR's ErUM-Data programme. Together with researchers from eleven different institutions, including DESY, Forschungszentrum Jülich, TUM, and the Leibniz-Institut für Astrophysik Potsdam, project coordinator PD Dr. Tim Ruhe (Lamarr, Physics) and Lamarr PIs Prof. Dr. Lucie Flek (Lamarr, NLP) and Prof. Dr. Dr. Wolfgang Rhode (Lamarr, Physics) are combining state-of-the-art developments in LLMs with research questions in physics.

DeepScenario develops spatial intelligence. AI that understands the spatial complexity of the real world to power intelligent systems in dynamic environments. Its technology turns video streams from any standard camera into a precise understanding of space, motion, and interaction, extracting 3D structure and context with high precision. The company delivers tailored solutions for defense and automotive, enabling next-generation autonomy and faster, more cost-efficient autonomous driving development. Customers and partners include BMW, Bosch, Mercedes-Benz, Volkswagen, and Continental, with strong roots in the German deep-tech ecosystem.

RobCo, founded in Munich in 2020, is a Physical AI-driven robotics company bringing autonomy into industrial operations. Its Autonomous Manufacturing Platform combines modular robotic hardware with an AI software stack, enabling fast deployment, continuous learning, and application-specific performance across workflows such as machine tending, palletizing, dispensing, and welding. Delivered through a robotics-as-a-service model with zero upfront investment, RobCo helps manufacturers automate manual tasks while minimizing complexity and risk. Backed by Lightspeed and Sequoia, the company operates across Europe and the US.

grubel is a research lab automating the specialization loop for critical knowledge work, starting with legal. Legal work takes specialization to the extreme. Each matter is its own world of facts, parties, customs, jurisdiction, and courts. Today, forward-deployed engineers sit with domain experts to build special-purpose systems, one problem at a time. It works. But it doesn’t scale, because it involves human expert labor at every turn: run a prototype, inspect its failures, make changes, try again. grubel is built on a simple bet: that loop can be automated. Our vision is a future where every matter builds its own AI, one that knows its world and does the best possible work within it.

Ecologic Computing rethinks information, computation, and their ecological impact for the digital age. The company investigates meaning-aware, task-dependent models of information transfer, treating information as more than just bits. Its work explores non-classical computing paradigms, including analog, hybrid, neuromorphic, and emergent computational systems inspired by nature and physics, extending beyond conventional digital logic. At its core is the idea of “ecological intelligence”: designing computing systems and interfaces optimized not merely for performance, but also for ecological impact, adaptability, systemic efficiency, and responsible integration into society. 

Portabiles HealthCare Technologies is a German digital health company improving quality of life for people with Parkinson's disease. Its patented, CE-marked and FDA-registered sensor technology enables AI-supported gait analysis for advanced patient care, therapy planning, and remote patient monitoring in clinical studies. The flagship offering, ParkinsonGo™, is an innovative care concept combining wearable sensors, digital monitoring, and expert counseling for patients and relatives, reimbursed by selected German health insurers since January 2024. Beyond Parkinson's, Portabiles aims to extend its technology to support further therapies.

As a national data competence center (DKZ), Come2Data aims to impart practical data skills—primarily for research and teaching, but also for public administration, the general public, and, in the future, the business sector. Our initial focus is on Saxony as a hub for science and research. Our goal is to offer a multi-tiered support system for solving data problems, to strengthen in-house expertise in FDM, and to create a network that brings researchers together with one another and with data experts. We support data-driven research and teach skills for competent data handling.

The School of Embedded Composite Artificial Intelligence (SECAI) is one of the three DAAD-funded Zuse Schools that foster education and research in artificial intelligence (AI) in Germany. The concept put forward by the SECAI host universities TU Dresden and Leipzig University is rooted in the tight integration of university studies, research, and applications, which will open up many career paths for young talents to successfully use and advance AI for the benefit of society and economy.

Physical AI is the next frontier of AI research. However, robot learning is bottlenecked by the cost of physical interaction. KE:SAI’s mission is to advance the efficiency frontier of robust and safe physical AI through fully open and reproducible research. KE:SAI – Kyutai ELLIS Scalable Autonomous Intelligence – is a non-profit frontier AI research lab co-located in Tübingen and Paris, dedicated to democratizing physical AI research by developing fully open stacks together with the research community. 

Personalized education is among the great unsolved frontiers of AI for the public good, and one of the highest-leverage. AIS is building the foundational infrastructure for AI-native learning at national scale: a sovereign, non-commercial platform commissioned by all 16 German federal states to bring adaptive intelligence into every classroom. At its core is an agentic tutoring architecture that fuses micro- and macro-adaptivity, knowledge tracing, and recommender systems into individualized learning trajectories — meeting each student where they are. Teachers are amplified by AI co-pilots for lesson design, material creation, and learning analytics, while open interfaces invite publishers and researchers to co-develop the next generation of adaptive learning resources. AIS aims to redefine what equitable, AI-powered education looks like for millions of learners.

grubel is a research lab automating the specialization loop for critical knowledge work, starting with legal. Legal work takes specialization to the extreme. Each matter is its own world of facts, parties, customs, jurisdiction, and courts. Today, forward-deployed engineers sit with domain experts to build special-purpose systems, one problem at a time. It works. But it doesn’t scale, because it involves human expert labor at every turn: run a prototype, inspect its failures, make changes, try again. grubel is built on a simple bet: that loop can be automated. Our vision is a future where every matter builds its own AI, one that knows its world and does the best possible work within it.