local inference · open-weight models · programme-controlled compute
Scientific programmes increasingly rely on AI to interpret instruments, reconstruct environments and coordinate complex workflows. Solpera operates its own locally managed compute infrastructure in Hobart so models, data and experimental processes can remain under programme control.
Our four-node NVIDIA DGX Spark system supports open-weight models up to 400B parameters and provides a common inference layer for spectral analysis, 3D reconstruction, agentic workflows and field-system development. Operating the stack locally allows each workflow to be designed around the instrument, governed as part of the programme and run without placing an external inference provider in the critical path.
The same architecture can be adapted for laboratories, institutions and companies requiring locally controlled AI, from a single workstation to multi-node systems.
| infrastructure | four-node NVIDIA DGX Spark system · Hobart |
|---|---|
| research workloads | spectral analysis · 3D reconstruction · multimodal inference · agentic workflows |
| operating model | local data and inference · open-weight models · no external inference required |
| deployment range | edge hardware · workstations · multi-node systems |
Explore locally controlled AI systems for external programmes →