CS & AI · University of Edinburgh · Healthcare AI & Drug Discovery
Third-year CS & AI student at the University of Edinburgh. Portfolio spans the full early drug development pipeline: multi-label QSAR adverse effect prediction (ChEMBL, SIDER, Morgan fingerprints, SHAP), ODE-based tumour resistance simulation, and conditional VAE for inverse molecular generation. DAAD RISE scholar (publication pending). Passionate about building software that accelerates evidence generation and improves health outcomes.
Machine learning analysis of 26,278 drug target-disease pairs, quantifying how strongly genetic evidence predicts successful clinical progression. XGBoost classifier with SHAP interpretability decomposes which types of genetic evidence matter most and for which disease areas.
OR = 3.25 (95% CI 2.79–3.79) · 26,278 pairs · Open Targets + ChEMBL
Unsupervised MOFA2 analysis across 11,060 TCGA samples spanning 31 cancer types identifies a conserved CEACAM5/B-cell exclusion programme, independently prognostic in the discovery cohort. Dissociated Cox modelling separates tissue-specific mechanistic drivers; external validation in an independent cohort (METABRIC) reported honestly, including a negative prognostic replication result.
End-to-end drug adverse effect prediction system for early-stage drug safety screening. Multi-label classifier trained on ChEMBL + SIDER across 15 adverse effect labels. Molecular featurisation using Morgan fingerprints (r2-r5) and 14 RDKit physicochemical descriptors. Weighted loss for class imbalance; atom-level SHAP interpretability for clinical explainability. Full-stack deployment with Flask REST API and JavaScript frontend.
15 AE labels · SHAP interpretability · Flask + JS
Four-compartment ODE model (sensitive, resistance A, resistance B, dual-resistant) modelling acquired resistance dynamics relevant to clinical treatment planning. RK4 numerical integration; four clinical dosing schedules (continuous, pulsed, metronomic, escalating); real-time growth/death flux analysis and resistance dynamics visualisation. Full-stack web application with interactive parameter controls.
Conditional GNN-VAE for inverse molecular design: generates candidate molecules conditioned on a target side-effect profile, closing the loop with the QSAR predictor (structure→effect and effect→structure). Systematic conditioning validation across 8 side-effect classes. Dataset expansion via DrugCentral in progress to improve generation quality.
Clinical trial recruitment risk predictor trained on 80,000+ ClinicalTrials.gov records. XGBoost classifier with SHAP interpretability surfaces which trial design factors (eligibility complexity, site count, condition rarity, sponsor type) most increase recruitment risk. Deployed as an interactive dashboard for pre-launch risk assessment.
Python library for multi-omics data harmonisation — solving recurring ID/ontology mapping, sample intersection, and missing-data problems shared across the projects above. Installable, tested, published on PyPI.
Multimodal AI system detecting manipulated politician videos. XceptionNet/EfficientNet for visual artefact detection (89-98% accuracy), Wav2Vec2 for audio deepfake detection (93-98% accuracy). Late fusion meta-learner combining modalities; target 96-98% combined accuracy.
Web-based ASL gesture recognition; ensemble of Random Forest (static) and LSTM (dynamic); 21-point landmark feature engineering with full WebSocket streaming.
Real-time hate speech detection model + Chrome extension for automatic comment blurring; 92% precision on test dataset.
I'm a third-year Computer Science and Artificial Intelligence student at the University of Edinburgh with a focused portfolio in healthcare AI and computational drug discovery. My projects span the full early drug development pipeline: multilabel QSAR adverse effect prediction, ODE-based tumour resistance simulation, and conditional VAE for inverse molecular generation.
I complement this with a self-directed biomedical informatics curriculum covering biological APIs, sequence analysis, multi-omics, and biomedical NLP. I have two sole-first-author manuscripts currently in peer review (PLOS Genetics; npj Precision Oncology), am a DAAD RISE research scholar, and have production full-stack software deployment experience.
I speak English, Norwegian, and German. Looking for opportunities in AI/ML for healthcare, drug discovery, or biomedical AI research.