ML Engineer · Neurodiagnostic + FinTech
I build and validate ML systems from data preparation and modelling through evaluation, explainability, and deployment — specialising in CT/MRI neuroimaging (stroke, dementia, brain tumour) and AI-powered FinTech automation.
Years Experience
Years in Data Science
Publications
I am a Machine Learning / AI Engineer focused on computer vision and applied AI, with experience delivering end-to-end ML systems from data preparation and modelling through evaluation, explainability, and deployment. My core work is in medical imaging AI — developing, validating and interpreting CNN-based classifiers for CT/MRI neuroimaging across stroke, dementia and brain tumour use cases at GoodFolio's GoodMind platform.
I have a strong Python background covering feature engineering, supervised learning, model validation, and interpretability (SHAP, Grad-CAM). I combine hands-on ML engineering with technical product thinking — translating user and business needs into reliable, measurable AI features — and have deployed AI agents using Google ADK, Vertex AI, and Cloud Run for regulated FinTech and RegTech environments.
End-to-end ML systems across medical imaging AI, bioinformatics, and AI-powered automation — from data curation and modelling through explainability and cloud deployment.
Medical Imaging
Medical Imaging
Multi-class CT classification distinguishing normal scans, ischemic stroke and haemorrhagic/bleeding cases to support urgent neuroimaging triage — validated with cross-dataset testing.
Medical Imaging
Hybrid deep learning framework on chest CT slices across four severity classes — Healthy, Mild, Moderate, Severe — with 5-fold cross-validation. Combined U-Net segmentation, CNN-based classification (VGG16, ResNet50, Xception), and classical ML (SVM, Gradient Boosting) with CLAHE preprocessing. Introduced a novel Brightness Intensity Range (BIR) method for pixel-level lung analysis, achieving high segmentation accuracy and fast hybrid-SVM inference compared to standalone CNNs.
View Publication
Bioinformatics
First study to profile oxytocin-pathway lncRNAs alongside target genes in an IDC breast cancer cohort, comparing expression in tumour versus matched normal tissue via RT-qPCR. Built and compared predictive models — Bayesian GLM, GLM, and LDA — evaluated through cross-validation, with the combined model achieving strong diagnostic performance. Led the ROC/AUC modelling workflow, produced all figures and tables, and contributed to peer-review revision.
View on GitHub
Bioinformatics
Predictive ML pipeline integrating ENCODE eCLIP binding signals from 139 RNA-Binding Proteins with Ribo-seq data to forecast treatment-driven translational changes in K562 leukaemia cells. Built and compared multiple classifiers across three iterative experiments — with hyperparameter optimisation via RandomizedSearchCV — and applied SHAP for biological interpretability of the best-performing model.
View on GitHub
AI Agents & FinTech
Developed and deployed AI agents for automated client outreach and RegTech inspection — Google ADK, Gemini, Vertex AI Agent Engine, Cloud Run. Explainable, auditable outputs for regulated environments.
AI Agents & FinTech
Research on enhancing banking customer experience through AI-powered invisible marketing — adaptive engagement strategies with measurable CX uplift in FinTech.
View PublicationFull list on Google Scholar.
Heatmaps are a starting point, not a validation strategy. After building stroke and dementia classifiers at GoodMind, I learned that real clinical trust requires cross-dataset testing, confusion-matrix analysis across subtypes, and documentation that clinicians can actually read — not just coloured overlays.
Most model failures in medical imaging happen before training starts. Building ML-ready cohorts from raw DICOM scans means getting clinical labels, metadata alignment, outcome proxies, and class balance right first — or your evaluation metrics will look great on paper and fail in deployment.
After deploying AI agents for RegTech and FinTech outreach on Vertex AI and Cloud Run, the biggest lesson was that auditability is a product requirement, not an afterthought. ADK handles orchestration well — but explainable, traceable outputs require deliberate design from the first prompt.
More recommendations: View LinkedIn profile.
Computer vision and medical imaging AI — specifically building, validating and explaining CNN-based classifiers for CT/MRI neuroimaging (stroke, dementia, brain tumour). I also build ML pipelines for bioinformatics and deploy AI agents for FinTech/RegTech automation on Google Cloud.
Yes. I work across the full pipeline — from dataset curation and model development through evaluation, explainability (SHAP, Grad-CAM) and cloud deployment. I also combine this with technical product management, translating clinical and business requirements into pilot-ready AI workflows.
Yes. I am open to ML engineering roles, research collaborations, and applied AI projects — particularly in medical imaging, diagnostics, or regulated AI. Feel free to get in touch.
Open to ML engineering roles, research collaborations, and applied AI discussions.
You may use the form below or contact me directly via email above.