§ 00Frontispiece
Senior Data Scientist · Mathematician · Tel Aviv ↔ Guangdong
A mathematician,at work.
I translate theorems into machines that learn. Two decades between a blackboard and a terminal — refinement schemes, geometric neural networks, anomaly detection, trustworthy AI. The curve above is a four-point interpolatory subdivision scheme — the subject of my doctorate, redrawn live each frame. Move your cursor.
“From mathematical principles to effective programming solutions.”
Most software is built on top of mathematics it never names. I work the other direction — beginning with the geometry, the inequality, the proof, then asking what useful behaviour falls out. The result tends to be smaller, stranger, and harder to break.
- · 01
Theory
Refinement schemes, convex geometry, statistical modelling — the toolkit of a working mathematician.
- · 02
Practice
Production data science across cyber, finance, ad-tech, automotive and life sciences. Fifteen years of shipped systems.
- · 03
Teaching
Lecturer at Technion (Guangdong) and HIT, head of AI training at Cyberpro. Research only matters if it transfers.
Between the blackboard and the terminal.
I hold a Ph.D. in Applied Mathematics from the Technion, where my dissertation introduced two generalisations of linear point-refining schemes — one constructing surfaces by refining non-intersecting curves, the other generating matrix-valued functions from sequences of symmetric positive-definite matrices.
For more than a decade since, I have moved between roles where rigour matters: cyber threat detection on the CAN bus, liquid-biopsy algorithms, log-analysis with Intel, prediction and pacing for ad-tech, and trustworthy AI for finance.
I currently work as Data Science Specialist at Fetcherr, lecture in the mathematics department of GTIIT, and teach Machine Learning for Cyber at the Holon Institute of Technology. Native English and Hebrew. Available for selected consulting engagements and public lectures.
- · Python
- · Statistical research
- · Deep learning
- · Geometric NN
- · Anomaly detection
- · Trustworthy AI
- · Pandas / NumPy
- · Public lectures
A career, refined level by level.
Each stop is a control point. The line is a four-point subdivision curve drawn through them — hover any stop to see the curve pull toward it.
- 2026 →Tel Aviv · full-time · currentFetcherrData Science Specialist
- 2024 — 26Guangdong, China · on-siteGTIIT (Technion · Guangdong)Lecturer, Mathematics
- 2025 — 26End-to-end cyber + GenAI programCyberproHead of AI training
- 2023 →Holon Institute of TechnologyLecturer · ML for Cyber
- 2023GNNs for fluid dynamics PDEsOrt Braude KarmielGeometric NN researcher
- 2021 — 23Trustworthy & explainable AI · financeTRSTaiSenior Data Scientist
- 2020 — 21Hardware-failure detection from logsQualitest × IntelSenior Data Scientist
- 2019 — 20CAN-bus cyber-attack detectionSafeRide TechnologiesData Scientist
- 2017 — 19Behavioural prediction from emailKnowmailData Scientist
- 2015 — 17Liquid-biopsy startup · HaifaSavicellHead of algorithm & ML
- 2014 — 15Pacing algorithms · ad-techKenshooResearcher
- 2010 — 14Derivatives & securitiesAlgorithmic-trading startupsAlgorithm Developer
- 2007 — 13Refinement schemes for geometric objectsTechnionPh.D., Applied Mathematics
Notes from the margins.
Selected essays and peer-reviewed work. The essays live on Medium — long-form thinking on innovation, physics, and the cognitive traps of expertise.
Essays · medium.com/@uriitai
Selected publications
- PaperApr 2025
Tighten the Lasso: A Convex-Hull Volume-based Anomaly Detection Method
A novel anomaly detection algorithm exploiting the convex-hull volume of a dataset; competitive with seven SOTA methods across ten datasets.
- PaperOct 2022
Parametric PDF for Goodness of Fit
A threshold-free framework for goodness of fit in classification — preserving the curve information typically lost to a fixed cut-off.
- Paper2013
Generalised refinement schemes for surfaces and SPD-matrix-valued functions
Doctoral dissertation, Technion. Two extensions of linear point-refining schemes to geometric objects.
Begin a conversation.
I take on a small number of consulting engagements, advisory roles, public lectures and graduate-level collaborations each year. Pick a thirty-minute window below — or write directly. I read every message.
A note, then.
If a calendar feels presumptuous — write a paragraph instead. I read every message and reply within a few days, in English or Hebrew.