Applied AI for systems where the physics matters
AGIgent is a deep-tech AI company. We work on a narrow set of problems — the ones where domain physics, sparse data, and operational constraints rule out generic AI tools, and where a working system has to be delivered, not just demonstrated.
The methods we apply are physics-informed neural networks, simulation-to-real pipelines, inverse-problem solvers, and goal-driven autonomous agents.
The sectors we serve today are space, defense, energy, and science-based R&D.
What we build
Four AI capabilities, applicable beyond the projects that first prompted them.
Physics-Informed Modeling
Neural networks that carry the laws of the domain inside the model itself.
Predictions stay within physically possible solutions, generalization improves, and the labeled-data requirement falls — because the model has been taught the system, not just the data.
Simulation-to-Real Pipelines
Training in three phases — pre-training on physics-grade synthetic data, fine-tuning on hardware-in-the-loop measurements, transfer learning onto real units.
The approach makes AI deployment possible where real data is scarce, expensive, or simply not yet available.
Inverse Problem Solving
Recovering the full state of a system from sparse measurements, or inferring the design that would produce a desired output.
Where classical iterative methods are slow or fail to converge, neural networks return an answer in a single forward pass — calibrated, traceable, and engineered to run in production.
Goal-Driven Agents
GoalDA, our agent framework for AI that holds an objective over long horizons.
Persistent goal coherence, proactive gap detection, contradiction resolution across heterogeneous sources, protected manual override with memory, full reference traceability.
Built for work that spans more than one session.
Space
AI for spacecraft, sensors, and the ground systems that depend on them.
The space industry is in the middle of a quiet shift. Constellations are growing too fast for ground teams to operate manually, payload data outruns the bandwidth available to bring it home, and the next generation of sensors reaches sensitivities that classical signal processing was not designed to recover. AI is moving on board the spacecraft, into the payload, and into the operational decisions of the agencies and operators that depend on them.
Today our work happens on the ground — on the AI that the next generation of space sensors, spacecraft, and ground operations will depend on. The instrument — models that rebuild signals hardware alone cannot deliver. The spacecraft — AI for autonomous tasking and fault response. The ground — fusion of sensor data, anomaly histories, and mission models into decisions worth making. Each line of work is designed for the constraints that will define it: radiation, size, weight, power, and the cost of being wrong in orbit.
An example of our work in this sector — ML4QS: AI denoising and signal recovery for quantum magnetometer payloads, with RAL Space (STFC, UKRI) as scientific partner.
Science-based R&D
AI for the instruments, experiments, and discovery pipelines of European research.
Some of the AI problems that matter most sit upstream of any operational deployment — in laboratories where the next class of scientific instruments is being designed, in research collaborations where the data outruns the methods used to analyse it, and in discovery pipelines where AI is moving from analysis tool to inference engine. The institutions doing this work — research institutes, universities, scientific instrument makers — need partners who can take a research question and return a system, not just a model.
Our work here functions as the laboratory bench of everything else we build. High-sensitivity sensor pipelines that later become space payload software. Real-time scientific image segmentation that later becomes industrial quality control. Inverse-design and explainable-AI methods that later become operational tools for grid and defense. We work directly with research partners on the problems they bring us, and carry the methods forward into our sectoral projects.
An example of our work in this sector — real-time AI segmentation and classification of holographic microscope imagery, delivered for the HoloZcan project.
Energy
AI for the grid, the renewable fleet, and the operators in between.
Europe’s power system is being asked to do something it was not designed for: integrate hundreds of gigawatts of variable, weather-driven generation while remaining balanced, stable, and within thermal limits. The forecasts, capacity ratings, congestion tools, and balancing-market workflows it depends on were built for a slower, more predictable world. Across TSOs, DSOs, balance responsible parties, renewable producers, and the aggregators between them, the same question keeps coming back — can AI close the gap between the system we have and the system the transition demands?
Our work is to help close that gap. Satellite Earth observation fused with grid physics for sub-kilometer weather and generation forecasting. Sensorless predictive Dynamic Line Rating that unlocks capacity from infrastructure that already exists. Imbalance economics turned from a penalty into a managed risk. Predictive congestion management at the TSO–DSO seam. And, at the industrial layer, AI that aligns production schedules with on-site renewable generation — turning a factory’s solar fleet from a sustainability line item into an operational asset.
An example of our work in this sector — GRIDFOR: sensorless predictive Dynamic Line Rating and hyperlocal renewable generation forecasting, in the framework of ESA Business Applications (Kick Start).
Defense
AI for the systems, sensors, and decisions
Defense in Europe is being reshaped by three simultaneous pressures. The systems being defended — networks, infrastructure, fleets, airspace — operate in increasingly contested electromagnetic, cyber, and autonomous environments. The pace of acquisition and modernisation is accelerating, demanding shorter design cycles and faster qualification. And the volume of sensor and intelligence data being generated has outrun the analyst capacity available to make sense of it.
We contribute at three layers. The component — AI that compresses calibration, inverse design, and qualification for RF, sensor, and EW hardware. The platform — machine learning for sensor fusion, autonomous tasking, and predictive health. The operational picture — AI that helps an operator see what classical methods miss, without taking the decision out of their hands.
Our visible portfolio today is in RF and antennas. The methods reach further.
From research to operation
Our work begins in research. It does not stop there.
Research
- frames the question
- selects the method
- defines the metrics of success
Feasibility
- delivers to the TRL scoped for the activity
- documents to agency or prime standards
- prepares the route to the next stage
Pilot
- integrates into the operational bus
- runs alongside the system it will eventually replace
- closes the sim-to-real gap on real units
Production
- delivers with monitoring and traceability
- transfers ownership to the receiving team
- supports through the operational lifetime
Our Founders
László BALÁZS
30+ years of AI research
Over 30 years of experience in research, AI development, and software engineering. László has co-founded multiple technology ventures and has expertise in neural networks, sensor-based systems, and technical leadership.
“Lorem ipsum dolor sit amet, consec
tetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua enim ad minim.”
Sebestyén DÓRA
20+ years of business development
Skilled in management, C++, Computer Vision, Video Analytics, and business development with a history of leading successful startups and innovative projects. Sebestyen has secured several awards for entrepreneurship and spearheaded groundbreaking research in wireless sensor networks.
“Lorem ipsum dolor sit amet, consec
tetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua enim ad minim.”
Judit VARGA
20+ years in product development
Over 20 years of experience in research, innovation, and product strategy. Judit has been part of 100+ product development projects and specializes in turning AI technologies into business solutions. Her expertise includes human-centered design, circular economy strategies, and leading zero-to-one product development.
“Lorem ipsum dolor sit amet, consec
tetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua enim ad minim.”
Projects
Our Partners
