Arine's platform and algorithms enable teams to identify and proactively address medication-related problems before they lead to costly interventions.

Led development of AI features into Veris's oncology-focused remote patient monitoring (RPM) platform. ● Developed an AI system to automate clinical progress note generation ● Established a prioritized roadmap, workflows, and standards to research, develop, and deploy science-based, FDA-compliant, data-driven capabilities into the RPM software platform.● Influenced decision-making around procurement and ingestion of electronic health record data for observational research, including de-identification, data modeling, and NLP methods.

Applied AI research for healthcare innovation. ● Collected annotations then trained Deep Learning models to segment cancerous lesions from 3D whole-body PSMA PET/CT scans from scratch.● Estimated the causal effect of diagnostic PSMA PET scans on patients’ therapeutic management and outcomes using Real World EHR data through a retrospective matched cohort study.● Developed statistical models to identify imaging-based biomarkers associated with patients' response to radioligand therapy.Collaboration with stakeholders from Janssen Data Science, UC Berkeley's Institute for Data Science, and UCSF Radiology & Urology. Joint appointment at UCB & UCSF.

Tech Lead - small teams building fully-automated AI-based flood-mapping tools and other commercial geospatial software products.

Timeseries analysis and Computer Vision to identify patterns and anomalies from satellite imagery ● Used unsupervised learning to characterize patterns and anomalies in geospatial time series datasets.● Developed a Computer Vision algorithm to identify wildfire extent in satellite imagery.● Project manager for four engineers to develop data pipelines, RESTful APIs, and Deep Learning-based models; communicated methods and key insights to clients through presentations and written reports.

Timeseries forecasting modeling for quant hedge fund customers● Iteratively improved continuous-learning predictive models of consumer traffic derived from millions of satellite image-derived car count estimates in terms of correlation to ground-truth data.● Created an XGBoost-based model to forecast time series of consumer sales with high accuracy and precision that used rigorous model selection through rolling-forward cross validation.● Developed an NLP pipeline to continuously extract, clean, and populate a database with ~260,000 areas of interest and their metadata -- a total of 2+ million data assets.

Researched novel methods to image, reconstruct, and co-align two 3D tomographic reconstructions from the same biological specimen. Applied these methods to derive insights into cellular functions.● Created a computational toolbox to measure properties of cellular features visible in 3D reconstructions and an automated pipeline to analyze ~1000 1-GB datasets.● Funded research to apply novel computational approaches to understand the relationship between nuclear architecture and epigenetics via a UCSF Program for Breakthrough Biomedical Research Award.● Published methodologies and scientific results in 10 peer-reviewed articles; presented at national conferences; received press coverage in The New York Times, Discovery, and others.

Dissertation: "Kinetics of SNARE-Driven Membrane Fusion in Reconstituted Single-Vesicle Fluorescence Resonance Energy Transfer-Based Assays".● Designed and conducted experiments and simulations to characterize the kinetics of biomolecular interactions on the millisecond timescale and 10s of nanometer length scale.● Developed software to identify 1000s of bimolecular interaction events from microscopy videos and quantify their relevant physical properties as a function of time.

Independent research in a large, productive structural biology lab.● Optimized conditions to overexpress, purify, and crystallize a protein-DNA complex. ● Collected preliminary x-ray diffraction data.