Software & AI Services

Custom ML model development

We design, train, evaluate, and deploy machine-learning models against client-provided datasets or proprietary data generated during the engagement. We work across supervised, self-supervised, and generative paradigms, with particular depth in structured scientific data (spectra, molecular graphs, simulation trajectories, time-series sensor data).

Deliverables include: trained model weights, training and evaluation code, a validation report (including out-of-distribution behavior and uncertainty quantification), and deployment instructions for the client's chosen platform.

Data pipelines & MLOps

Scientific data rarely arrives clean. We design and implement ETL pipelines, feature engineering modules, and data quality frameworks that handle the irregularity of laboratory and instrument data. For clients moving to production-scale inference, we set up model registries, versioning, monitoring, and automated retraining pipelines using open-source MLOps tooling (MLflow, DVC, Prefect) on client-controlled infrastructure.

We do not require clients to use cloud-specific services; all pipelines are designed to run on the client's existing compute, including on-premises HPC.

AI advisory & audits

Independent technical review of existing AI/ML systems: model architecture, training methodology, data provenance, evaluation metrics, and deployment risk. We deliver a structured audit report that identifies issues with reproducibility, distributional shift, fairness implications, or regulatory compliance (particularly relevant for clients in regulated industries). Advisory retainers are available for teams that want ongoing technical input during model development.

Scientific simulation software

We develop, extend, and maintain simulation software for materials science, fluid dynamics, structural mechanics, and biophysics applications. Projects range from porting existing research code to a production-grade library, to designing new simulation frameworks for bespoke physical systems. All software is delivered with unit tests, integration tests, API documentation, and a usage example suite.

Languages: Python, C++17, Fortran (legacy extension), Julia. GPU kernels in CUDA and ROCm where performance-critical.