An AI leader with 15+ years of experience across startups, pharma, and academic research. Specializing in building cross-functional teams and translating AI research into production-grade systems.
Led applied AI programs from scientific framing to validated models, integrating multi-omics, electronic health records (EHR), and imaging to uncover novel disease mechanisms.
Led AI and big data initiatives for clinical trials within Specialty Medicine R&D.
Built NLP and ML-based optimization engines for large-scale production platforms.
Led large-scale software engineering programs and enterprise architecture initiatives
Large-scale computer vision platform for organelle phenotyping in patient-derived iPSC neurons, leveraging Vision Transformers (ViT) trained on fluorescence microscopy images. The project proposes contrastive, perturbation-aware learning under weak labels to quantify phenotypic effects across diverse genetic and chemical perturbations at the scale of tens of millions of single cells.
Led some of the first digital clinical trial efforts in Israel, integrating smartwatch signals and eDiary in movement disorders, and deploying smart inhaler programs for COPD and asthma.
Predictive models to estimate patients’ propensity for early trial termination, in Israel's largest pharma company.
ML pipelines integrating genomics, transcriptomics, metabolomics, proteomics, and clinical data to identify candidate biomarkers for diagnosis and prognosis, including my discovery of miR-181 as a prognostic biomarker for ALS.
A convolutional neural network for sleep and wake detection from wrist actigraphy, showing improved performance over traditional rule-based approaches and representing an early deep learning application to wearable time series.
A target trial emulation and causal inference methods to large-scale Clalit electronic health records, enabling principled estimation of treatment effects from observational data using the potential outcomes framework.
Co-developed a clinically validated ML model for forecasting advanced liver fibrosis and cirrhosis within five years using routine blood tests. Prospectively validated on real patients and shown to outperform standard screening (FIB-4), identifying high-risk individuals who would otherwise go undetected.
Developed an unsupervised machine learning framework for risk scoring and profiling of user cyber behavior, focusing on detection of risky browsing patterns without labeled data. The work proposed a feedback-based learning scheme and was evaluated on real organizational settings.
Full and up-to-date list available on Google Scholar: https://scholar.google.com/citations?user=zOessRsAAAAJ&hl=en