TLDR. PhD in AI and Cancer Imaging at the Institute of Cancer Research; co-created OsiriXgrpc, a clinical AI plugin used at The Royal Marsden Hospital for research. Previously delivered a 30% volume uplift and £5M increase in GWP on motor new business in commercial ML in the insurance industry. 11 abstracts/papers across MICCAI, AAAI, ISMRM, MIDL
I build AI systems across healthcare, insurance, and applied research. PhD in AI and Cancer Imaging from the Institute of Cancer Research, where I shipped a clinical AI platform now used at The Royal Marsden Hospital. Before research, I led ML-driven pricing at Ageas Insurance.
Now running an AI and data analytics consultancy taking projects from scoping through to production deployment.
- 🛠️ Building: End-to-end pricing engine for an insurance client, unifying technical (burn cost) and commercial (market price) models into a single optimiser pipeline.
- 🔬 Experimenting: Foundation-model adaptation for medical imaging - see
dinov2-small-medniston Hugging Face. - 📖 Reading: The Foundation Models Initiative by Stanford CRFM; Interpretable Machine Learning by Christoph Molnar; recent MICCAI proceedings.
|
clinical AI plugin Used at The Royal Marsden Hospital |
GWP uplift · 30% new-business volume Motor new business at Ageas via ML-optimised pricing |
abstracts / papers MICCAI · AAAI · ISMRM · MIDL CRUK 4-Year PhD Studentship |
| Domain | Detail |
|---|---|
| Healthcare AI | Co-created OsiriXgrpc - a gRPC plugin for OsiriX that brings real-time AI inference into radiologists' DICOM workflows. Used at The Royal Marsden Hospital, serving care for patients. Funded by the MedTech SuperConnector and the Sarcoma Accelerator Consortium. |
| Commercial ML | Led ML pricing at Ageas - delivered a 30% volume uplift and £5M increase in GWP on motor new business |
| Research | 11 abstracts/papers (MICCAI, AAAI, ISMRM, MIDL). Cancer Research UK 4-Year PhD Studentship |
| Open source | Contributor to The Turing Way (2.1k★) and NL-Augmenter (788★) |
"...the Medical Computer Vision Practical Tim produced has been used by multiple medical student cohorts to better understand what goes on under the hood in a computer vision classifier. I strongly recommend him for the person he is and the expertise he holds."
Nicholas Fuggle · Associate Professor of Rheumatology · Co-organiser, Alan Turing Institute Clinical AI Interest Group
Removes the historical bottleneck for clinical AI translation: state-of-the-art models live in Python, but radiologists work in OsiriX. The plugin lets the two talk in real time, without forcing the clinical team to leave their viewing environment. Source & documentation: github.com/osirixgrpc/osirixgrpc · osirixgrpc.github.io.
Interactive medical-image classification with GradCAM explainability visualisations - upload an image, see the model's classification and the regions it attended to. Built around DINOv2-small vision transformers fine-tuned on MedMNIST v2 and RSNA pneumonia. MONAI-compatible, SafeTensors, Apache-2.0.
Used as teaching material for:
- The Clinical AI Summer School at The Alan Turing Institute - the Institute's hands-on training programme for clinicians.
- The clinical student cohort at the University of Southampton.
Companion to a 10-chapter clinical AI curriculum for radiology fellows with no ML background (notebook, glossary, Gradio demo, fine-tuned model). The same explainability principles I worked on during my PhD on soft-tissue sarcoma at the Institute of Cancer Research, distilled into public artefacts clinicians can pick up in minutes.
Automated checks over sensitive tabular data: anonymise, generate synthetic data, then adversarially test what survives.
| Type | Title | What |
|---|---|---|
| Space | Synthetic Data Privacy Audit | Audit synthetic data with 4 privacy metrics + 10 re-identification attacks |
| Space | Privacy Lab | Anonymise a CSV, then red-team it with 10 attacks to see what leaks |
| Repo | data-anonymization-toolkit | The config-driven engine behind both: anonymisation, synthetic generation, adversarial validation |
| Type | Title | What |
|---|---|---|
| Site | The AI Loop | Interactive map of the circular economy of AI (FT edition) - click any company to trace who funds it, how much, and whether each deal is confirmed or merely reported |
| Site | Running AI Locally | Model + GPU + cost comparisons and a step-by-step Ollama + Claude Code setup guide |
| Type | Title | What |
|---|---|---|
| Repo | tabular-modelling-pipeline | 8 architectures (CatBoost, XGBoost, FT-Transformer, TabM, LocalGLMnet, DRN and more) with Optuna tuning, ensembling and SHAP/Captum interpretability |
| Models | bike-sharing · house-prices | Benchmarked tabular models with companion open datasets on Hugging Face |
| Space | Vision Extract | Pull structured tables out of images and video using Claude vision |
Pushing current models on domains I know deeply, for fun. A Slay the Spire codex (multimodal embeddings, an interactive UMAP map, a synergy inspector, and an LLM deck builder) and One Piece TCG tools (semantic search, auto deck builder), backed by 13 open datasets with a few hundred downloads between them.
Follow @t22000t for updates.
Tech I use most (click to expand)
ML & Deep Learning
Medical Imaging & CV
Languages
Infra & Data
- OsiriXgrpc: Rapid Development and Deployment of State-of-the-Art AI for Clinical Practice - AAAI 2022 (AI2SE Workshop)
- Radiomics Using Disentangled Latent Features from Deep Representation Learning in Soft-Tissue Sarcoma - MIDL 2023
- Multimodal Fusion for Radiogenomics Classification of Brain Tumor - MICCAI 2021 (BraTS Workshop)
- Uncertainty Quantification using U-Net with Monte Carlo Dropout - MICCAI 2021 (QUBIQ Workshop)
- Test-Retest Repeatability of Data-Driven Radiomic Features from Deep Learning - ISMRM 2022
Full list (11 papers): see LinkedIn Publications.
Open to conversations about applied AI roles, consulting engagements, and research collaborations where data and experimentation drive real commercial outcomes at speed.