Technology & infrastructure
Practical compute for industrial AI workloads.
A technical overview for infrastructure providers and GPU cloud partners, grounded in the needs of the workload rather than GPU usage for its own sake.
Workloads
Built for time-series intelligence.
The product concept combines conventional application services, high-volume sensor processing, and optional advanced model workloads.
Predictive analytics
Condition modeling and equipment risk estimation.
Time-series anomaly detection
Signal patterns, baselines, and multivariate deviations.
Sensor data processing
Validation, normalization, aggregation, and feature generation.
Vibration & deep learning
Time-series forecasting, vibration analysis, embeddings, and model evaluation.
Optional AI assistance
Potential LLM-powered maintenance search and support workflows.
Reference architecture
A modern, portable software stack.
The architecture can adapt to cloud, private cloud, or on-premises requirements as product validation and customer needs evolve.
Python + FastAPI
Model services and application APIs
PostgreSQL
Relational application records
PyTorch + scikit-learn
Anomaly detection & forecasting (PatchTST / TimesNet experiments)
TimescaleDB + Redis
High-volume telemetry and caching
Docker + Kubernetes
Portable, orchestrated services
MLflow + Grafana
Model tracking and observability
Llama / Qwen (optional)
LLM-assisted maintenance recommendations
React
Web application interface
AWS
Current application infrastructure; dedicated GPU providers under evaluation
GPU requirements
Optional acceleration, scaled by evidence.
GPU requirements depend on actual workloads and benchmarks. Basic sensor processing and many anomaly-detection tasks can run on CPUs. GPUs are not mandatory for the entire platform.
Potential GPU use
Advanced deep learning experiments, high-throughput model evaluation, representation learning, and optional language-model workloads.
Initial requirement
4 × H100 80GB
Initial requirement. Main GPU class: NVIDIA H100 80GB or H200 141GB. Preferred region: US West, ideally Northern California.
Estimated spend
$12–18K/mo
Estimated monthly GPU / AI spend. Actual needs depend on real workloads and benchmarks.
Preferred infrastructure
Reliable foundations for evaluation and scale.
- Dedicated GPU instances where needed
- Persistent, high-throughput storage
- Infrastructure and application monitoring
- Scalable CPU and GPU compute
- Secure networking and access controls
Deployment options
Cloud
Managed scale and rapid iteration.
Private cloud
Dedicated environments and policy control.
On-premises
Local operations for specific security and latency needs.