TSPmicro · Edge AI for predictive monitoring
Time-series prediction on microcontrollers.
Predictive monitoring that runs where the data is: on the sensor node, without cloud, with data that never leaves the site.
What we do
TSPmicro brings time-series forecasting to low-cost microcontrollers. A compact neural network of a few tens of kilobytes predicts how a machine or process will behave next and flags deviations before they become failures.
Edge-native
Runs on €10-class hardware. No cloud subscription, no continuous connectivity, no data transfer.
Physics-informed
Engineering knowledge plus data: models learn from a handful of failure examples instead of thousands.
Fleet-ready
Fleet-adaptive calibration with frozen weights: one model, tuned to each installation without retraining.
Applications
Industrial predictive monitoring for:
- heat-pump and HVAC fleets,
- energy and process plants,
- storage and drying in agriculture.
Technology and IP
TSPmicro builds on research at the Karlsruhe Institute of Technology (KIT) in turbulence, heat transfer and machine learning. The core method is the subject of a pending EP patent application on efficient time-series neural networks for edge devices, IP held by KIT.
Selected publications:
- I. Otić (2025). Optimized LSTM neural networks for material damage predictions. Nuclear Engineering and Design 442, 114264. doi:10.1016/j.nucengdes.2025.114264
- B. Günseren et al. (2026). Constructing a Reynolds stress model of decaying homogeneous isotropic turbulence using physics informed neural network. Journal of Computational Physics 553, 114725. doi:10.1016/j.jcp.2026.114725
- Full record: ORCID 0009-0003-8529-6858
Team
Dr. Ivan Otić, founder and inventor, has a long-standing research record in Computational Fluid Dynamics, turbulence research, nuclear engineering and design, and physics-informed machine learning.
Contact
ivan.otic@tspmicro.com
Karlsruhe, Germany
LinkedIn