AI across the medical research landscape

We build computer vision models for radiology and pathology research: organ and lesion segmentation, nodule detection, whole-slide image analysis, and quantitative imaging biomarkers. Models are validated on held-out and, where possible, external cohorts.

We turn unstructured medical text into research-ready variables: diagnoses, medications, dosages, adverse events, timelines. We also build retrieval-augmented LLM assistants based on guidelines, protocols and scientific literature, deployed on-premises when the data requires it.

We develop graph neural networks and other molecular models for activity, toxicity and ADMET prediction, virtual screening of compound libraries, and structure-aware candidate generation. All models are tuned to your assays and chemical space.

We analyse physiological time series from bedside monitors, Holter devices and wearables for arrhythmia detection, sleep staging, seizure detection and remote-monitoring endpoints in clinical studies.

We build interpretable models for readmission, complications, disease progression and treatment response, including survival analysis and causal inference methods for observational data, with calibration and fairness checks across patient subgroups.

We apply machine learning to high-dimensional omics data: biomarker discovery, patient subtyping, variant prioritisation, and integration of omics with clinical and imaging data for a fuller picture of disease.

We help sponsors and CROs use AI across the trial lifecycle: matching patients to eligibility criteria, forecasting recruitment, detecting anomalies in trial data, and automating parts of medical writing.

Many of the most interesting medical problems have no ready-made solution. We study the question, the data and the constraints, and design a custom architecture, from small models that run on a device to large multimodal systems.