Ehud Karavani

Email LinkedIn Scholar Medium GitHub

Highlights

• Machine learning researcher and causality expert with 9 years of experience in healthcare

  • Applied researcher and data science working solo, as tech-lead, or as squad leader

• EB1-A visa approved (US)

• 1st author Cell paper and co-inventor on 3 US patents

• Excellent Python coder

  • Creator of causallib - an open-source Python package for causal inference.
    800+ stars and 100+ forks on Github.

  • Developed high-throughput frameworks for quasi-experiments
    from statistical engines to dashboards exploring results and supporting decisions

• Received an IBM Research Accomplishment award (2023)

• Communicator: PyData conference speaker, lecturer, and podcast interviewee

Causal inference, machine learning, deep learning, statistics, data viz, communication, Python

About me

I bridge the gap between rigorous statistical learning and robust software engineering. I specialize in automating analyses into scalable frameworks. Whether it’s architecting the backend of a statistical engine or designing visualization-heavy dashboards, I close the loop between complex modeling, reusable tools and actionable stakeholder insights.

As a project leader, I translate vague business/research questions into concrete hypotheses, manage agile milestones, collaborate with international peers, and navigate the nuances between technical and non-technical communication.

A T-shaped integrative thinker. I thrive on cross-pollinating fields: I have adapted biostatistics concepts to tailor modern transformer-based deep-learning architectures for biological data, applied machine learning theories of invariance to promote fairness in genetic risk scores, and used formal causal graphs to deconfound learning from multiple sources and drastically improving model generalization. I’m also a passionate advocate of synthesizing engineering practices into research practices: academic workflows as git workflows, test-driven modeling, and Clean Code for research code.

Experience

2017 – present	Staff Machine Learning Researcher Causal Machine Learning for Healthcare and Life Science, IBM-Research Creator of causallib – a one-stop-shop open-source Python package for flexible causal inference modeling. Received an IBM Research Accomplishment award (2023) Client project leader from start to finish: eliciting information from domain experts, translating vague clinicians’ questions into concrete statistical hypotheses, answering them, and communicating the findings Led, designed, and engineered a reusable framework for drug discovery, applying high-throughput causal inference to observational healthcare data Managing a team of 5 researchers. Leading the scientific pipeline, system design, and visualization app Generating 100s of hypotheses in minutes Serving 4 external engagements with top pharma clients, bringing millions in revenue Individual Contributor (IC) Causal inference consultant for projects in the US, UK, France, Japan, Kenya, South Africa, and Switzerland Led global strategy at IBM Research for causality in drug discovery Steered research agenda and technical focus areas, reporting directly to Research VPs Oversaw research of subgroup discovery for adaptive experimentation using Bayesian inference Mentored 10+ students and interns Onboarding lead, onboarding 10+ researchers Teaching academics how to apply software development fundamentals to research, delivering maintainable production-grade research code Published 10+ papers and issued 3 US patents 2024: “GLM-ification” of deep learning models, bringing established concepts from biostatistics into transformer-based deep-learning LLM-like models, tailoring them for biology. Implemented concepts from generalized linear (mixed) models (e.g., zero-inflated negative-binomial regression or random effects) in PyTorch Deconfounded learning from multiple fragmented sources improving model generalization Identified and quantified data confounding bias (batch-like effects) leading to poor generalization in clients Deconfounded learning using approaches from domain adaptation, invariant risk minimization, and conditional autoencoders, drastically improving model generalization 2025: Quantum Advantage task force member: developing and testing quantum algorithms for combinatorial optimization problems Applied AI and analytical approaches to improve quantum algorithms, like finding better initial parameters, reducing variational optimization rounds hardware usage and saving costs
2022	Principle Statistician Laboratory for Gait & Neurodynamics, Ichilov Hospital Lead statistician in a clinical study Bayesian hierarchical/multilevel models and causal inference for gait analysis in multiple sclerosis patients Bayesian multilevel models for hurdle models of repeated patients’ measurements Formal causal inference with DAGs for minimizing inessential tests, saving over 3 hours of unnecessary tests by clinicians per patient.
2016 – 2017	Teaching Assitant The School of Computer Science, Hebrew University of Jerusalem Introduction to Data Science Workshop in Computational Bioskills
2015 – 2016	Research Associate / Computational Biologist Institue for Medical Research Israel-Canada, Hadassah Hospital Developed novel methodologies for finding high-resolution protein-RNA interactions using high-volume RNAseq data

Education

2016 – 2019	M.Sc. in Computer Science and Computational Biology Faculty of Science, the Hebrew University of Jerusalem, Israel Thesis: quantifying the utility of embryo selection using genomic prediction of traits published in Cell Predicting physical traits from DNA (GWAS) using classical, machine learning, and deep learning methods Pioneering the effects of prediction-based embryo selection in IVF
2013 – 2016	B.Sc. in Computer Science and Computational Biology Faculty of Science, the Hebrew University of Jerusalem, Israel Dean’s List of Academic Excellence (2016) Research scholarship from IMRIC (2016) Bachelor’s thesis published in Nucleic Acids Research

Outreach

• PyData speaker

• Causal Bandits podcast interviewee

• DataNights causality series lecturer

• Recurring DataHack mentor and judge

• Co-Organized the 2018 Atlantic Causal Inference Conference Data Challenge

• TDS Editor’s Pick on Medium

Skills

Programming skills	Python & its scientific and ML stack (fluent) Pandas, Polars, DuckDB, Ibis, Numpy, Scikit-Learn, PyGAM, Statsmodels, PyTorch (lightning), PyMC, Bambi, Arviz, Matplotlib, , Seaborn (objects), Altair, Streamlit, cvxpy, Pydantic, Hydra, Ray… R (when needed) SQL (but Ibis when possible) Git + GitHub Continuous development (Travis, GitHub Actions) Linux and remote development (Cloud/AWS + Jupyter lab / VS Code) Jupyter, Quarto, Latex, Typst
Technical skills	Causal Inference Machine Learning and Deep Learning Statistics and Bayesian Inference Data Visualization Verbal & written communication Programming, software engineering and development
Languages	Fluent English Native Hebrew
General	Data enthusiast Musician 🎸, hiker / backpacker 🏔️ | Friendly 🙂

Awards

2023	IBM-Research Accomplishment For my work on causallib and research engagement with the Cleveland Clinic Foundation
2019	Best of RSNA For the paper Predicting Breast Cancer by Applying Deep Learning to Linked Health Records and Mammograms, published in Radiology
2019	Best Talk: Israeli Population Genetics Meeting For the paper Screening Human Embryos for Polygenic Traits has Limited Utility
2019	Featured Theory of the issue (Cell) For the paper Screening Human Embryos for Polygenic Traits has Limited Utility
2016	Dean’s list of academic excellence

Publications

May go out of date. Please see my Google Scholar page for the most up-to-date information.