Eliezer Mashihov

Eliezer Mashihov

Data Scientist · Developer

I am a data engineering student and researcher working at the intersection of machine learning, mathematics, and applied systems. I focus on translating deep theoretical intuition into tangible engineering solutions.

About Me Projects Research Contact
GitHub LinkedIn CV ↓
scroll

Background

Who I am and what I'm currently working on.

I am a Data Engineering student and aspiring MSc candidate at the Technion, motivated by a desire to understand the full lifecycle of data. Whether I am building foundational data pipelines or extracting insights through advanced machine learning, my focus is on turning raw information into intelligent, automated systems. I operate best at the intersection of applied systems and theoretical research, where big ideas have to be proven in practice.

I approach my work with a meticulous, investigative mindset. The most rewarding part of engineering is the opportunity to step into unmapped domains, ask the right questions, and build a solution that leaves no stone unturned. I love the iterative grind of the research process—the cycle of designing an experiment, identifying the weaknesses, and engineering a system from the ground up until it is undeniably solid.

2022 – 2026

BSc Data Engineering

Technion — Israel Institute of Technology · Dean's Award of Excellence

2014 – 2017

Darski High School

Graduated with honours

2025 – Present

Research Assistant — Prof. Avi Gal & Prof. Assaf Shohar

Technion · Modeling legal statutes as database constraints (in collaboration with the University of Haifa), and representing neural networks as linear functions using classical linear algebra

2018 – 2020

Soldier & Medic — Givati Brigade

Israel Defense Forces · Developed methodical documentation, planning, and operational execution skills

Selected Work

Data science projects, ML experiments, and applications.

Alzheimer's Classification via Large Language Models GitHub ↗

Applied large language models to classify transcripts of people with Alzheimer's disease. Investigated the linguistic subtleties — specific patterns in speech and word choice — that distinguish Alzheimer's transcripts from healthy controls. Supervised by Prof. Roy Reichert.

LLM Memorization of Literary Texts GitHub ↗

Investigated whether large language models can memorize information from books and how this depends on the linguistic and literary properties of the source text. Found correlations between specific textual characteristics and a model's ability to recall book content. Supervised by Eyal Ben David.

GNN-Guided Heuristics for Classical Planning GitHub ↗

Integrated learned heuristics into classical planning problems using graph neural networks. The GNN is trained to predict the quality of planning states, guiding the planner toward more efficient search trajectories and reducing the number of states explored. Supervised by Prof. Erez Karpas.

Simplified Observation Models for Robot Decision-Making GitHub ↗

Explored computationally lightweight observation models for robot decision-making that remain accurate enough for reliable decisions. Analysed the error bounds introduced by model simplification and demonstrated that a robot can make sound decisions under simplified — yet bounded — uncertainty models. Supervised by Prof. Vadim Indelman.

Modeling Legal Statutes as Database Constraints

Research assistant project in collaboration with the University of Haifa. Models state laws as formal constraints on relational databases and builds an interface through which a database instance can be evaluated and repaired to comply with regulatory requirements. Explores the intersection of legal reasoning and database theory. Supervised by Prof. Avi Gal.

Neural Networks as Linear Functions via Classical Linear Algebra

Research assistant project building on Prof. Shohar's recent publication. Represents trained neural networks as linear functions using tools from classical linear algebra. The framework enables direct inspection of the features learned by a network, offering a new lens for neural network interpretability without requiring additional tooling. Supervised by Prof. Assaf Shohar.

Publications

Academic papers and writings.

Modeling Legal Statutes as Database Constraints Ongoing

Research Assistant · Technion — in collaboration with the University of Haifa

This project models state laws as formal constraints on relational databases. The core challenge is translating legal language — which is often ambiguous and context-dependent — into precise formal constraints that a database system can enforce and repair. The work sits at the intersection of legal reasoning, formal methods, and database theory.

Neural Networks as Linear Functions via Classical Linear Algebra Ongoing

Research Assistant · Technion

This work represents trained neural networks as linear functions using tools from classical linear algebra. Rather than treating a neural network as a black box, the framework exposes the network's learned features directly through algebraic decomposition. The goal is to investigate what a network has actually learned — and why — in a way that is interpretable, rigorous, and does not require additional tooling or probes.

Get in touch

Open to research collaborations, data science discussions, and new opportunities. Feel free to reach out.