Scientific code made to run on your cluster, your cloud or an air-gapped network.
Parallelisation, GPU kernels, simulation pipelines, testing and deployment for numerical software that has to run in production.
- 01Runs where you need it
- Workstation, cluster, cloud account or air-gapped network, from one reproducible build.
- 02Correct as it changes
- Regression tests against reference solutions, continuous integration and full provenance, so the code stays right after the person who wrote it has moved on.
- 03Fast where it counts
- Profiling first, then MPI decomposition and GPU kernels for the parts that dominate runtime.
01Capabilities
What we do
Solvers, pipelines, parallel code, deployment and tests.
- Solver and pipeline development
Purpose-built scientific software for your computational problem.
- Domain-specific solvers on FEniCSx, deal.II, PETSc and OpenFOAM
- Pipelines from CAD import through meshing, solving and post-processing
- Parameterised configurations for design exploration
- Integration with your engineering data and workflows
- HPC and parallel computing
From one node to a thousand cores without rewriting the physics.
- MPI domain decomposition for distributed runs
- GPU acceleration of the kernels that dominate runtime, in CUDA
- Profiling, bottleneck analysis and optimisation
- Cloud bursting with cost controls for peak studies
- Sovereign deployment
It runs where you need it, including networks with no internet access.
- Air-gapped and classified network installation
- On-premise, cloud and hybrid architectures
- No licence server, no vendor dependency for operation or updates
- Reproducible builds from source, on a clean machine
- Quality engineering for scientific code
The practices that keep numerical software correct while it changes.
- Continuous integration with regression tests against reference solutions
- Mesh convergence and solver verification frameworks
- Version-controlled workflows with full provenance
- Documentation and handover so the code outlives the engagement
- Language models on your infrastructure
Small, domain-tuned models that run inside your network.
- Small language models fine-tuned on your documents, standards and procedures
- Retrieval over engineering data with the source shown for every answer
- Deployed on-premise or air-gapped, on the GPUs you have
- Evaluated against your own questions before anyone relies on it
- 01Profiling report with the cost of each option
- 02Parallel or GPU implementation, verified against the original
- 03Test suite and continuous integration
- 04Container images and deployment scripts
- 05Documentation and handover
02Common problems
Problems we solve.
01Fourteen hours per case on one workstation.
We profile it and map the path to a cluster or a GPU, with the cost of each option, before writing a line.
02The researcher who wrote it has left.
Tests, documentation, a build that works on a clean machine, and handover to the team that has to maintain it.
03The network has no internet and the vendor cannot come in.
Reproducible builds from source, installed and updated inside the network, with no call-home.
04The general-purpose model cannot see the documents that matter.
A small language model tuned on your own material and run on your own hardware, with the source shown for every answer.
Tell us your pain points.