Remaining useful life prediction for turbomachinery
RUL models on the NASA C-MAPSS benchmark dataset: early detection of degradation dozens of cycles before failure. The method transfers to pumps, compressors and gas-pumping units.
Expertise
Expertise
An engineering team with expertise in machine learning, computer vision and industrial data. We prove our competence with open R&D projects on benchmark datasets — before asking for access to your data.
Stack
Technology stack
Machine learning for time series
Anomaly detectors (autoencoder, isolation forest), survival models for remaining useful life (RUL), gradient boosting, explainability (SHAP)
Computer vision
Detection and segmentation (YOLO family), tracking, edge inference on GPU (Jetson, on-prem servers), fine-tuning for site conditions
LLM and RAG
Local Qwen/Llama-class models, RAG pipelines with source citation, document AI (tables, P&ID), SSO and role-based access
Industrial data and integrations
OPC UA, Modbus, historians (PI, WinCC, PHD), CMMS (SAP PM, 1C), LIMS, SCADA/EMS — read-only, no writes to the control loop
Infrastructure
On-premise and private cloud within Kazakhstan's jurisdiction, Docker/Kubernetes, MLOps, high-load systems (10,000+ concurrent users)
R&D
R&D and capability demos
Open datasets are an honest way to show the method before an NDA is signed.
Virtual flow meter on Volve data
Open data from the real Volve field (Equinor): a hybrid of physics (VLP/IPR) and ML for continuous well-rate estimation without test separators.
PPE and danger-zone detection
Detectors fine-tuned on open datasets (SH17, CHV): helmet, vest, safety harness, entry into a danger zone — a live demo on industrial site footage.
RAG assistant for regulatory documentation
A local LLM with search over a corpus of Kazakhstan (ST RK) and GOST standards: answers cited down to the clause and document revision — entirely inside a closed perimeter.
Next step
Test us on your data
Give us an anonymized telemetry export with known failures and we will show you the precursors.
- 1A 30-minute call: the task, the data, who signs off on security
- 2Data audit or a retrospective analysis on an anonymised export
- 3An 8–12 week pilot with success criteria agreed before the start
NDA from the first contact
Reply within one business day