Tom Murarik
Senior Data Scientist
About
A modern decision scientist should think like a mathematician, code like a software engineer, deploy like a data engineer, and communicate like a business analyst. I build end-to-end optimisation systems with this reliable stack.
At Ocado Logistics I own projects across last mile van routing, forecasting, and network design, delivering millions in cost savings. I prioritise interpretability and maintainability over mathematical complexity, because a solution that cannot be deployed or explained is not a solution.
Experience
- Network Optimisation: solving highly constrained facility location problems using metaheuristic frameworks. With large-scale ensemble simulations, directly informed the decision of new Customer Fulfilment Center locations as part of Ocado's expansion strategy.
- Multi-Objective Optimisation: deciding optimal driver shifts by pairing this approach with Monte Carlo simulations leading to significant improvement in routing efficiency metrics and changes in driver hiring/headcount.
- VRPs: improving vehicle routing efficiency by enabling parallelisation and integrating cutting-edge new algorithms on the optimisation framework that routes the tens of thousands of vans used by the Ocado Group's retail partners.
- Time-series forecasting: deployed tree-based ML models across the business, applications include hourly sales and basket size forecasting. Automated manual methods and decreased forecast error by up to half vs manual Excel-based legacy methods.
- MLOps: created centralised ML forecasting pipeline templates, enabling easy repeated use across three separate retailers. Suitable for templating new forecasting projects across teams.
- Decision Model Deployment: deployed optimisation algorithms for use by analysts; included dynamic right-sizing of VM depending on problem size, easily accessible solution storage and Streamlit UI.
- Cloud: ownership of business-critical pipelines, including hourly refreshes of key forecasts and the cloud cost vs benefit trade-off of running these models. Extensively involved in set-up and architecture of GCP eco-system.
- SWE: development of heuristic architectures in Scala for a SoTA routing solver trusted by multi-national haulage companies such as Kuehne + Nagel.
- Heuristics: improved optimiser performance while decreasing run-time, presented at the VeRoLog conference Jun 2025.
- DNNs: work on PyTorch models that predict loading times for trucks depending on the content of the load, providing better bounds for the overlying optimisation model.
- Stakeholder engagement: secured new business from cash-and-carry providers at industry conferences.
- Agile work in interdisciplinary teams; taking mathematical concepts from client spec to development to deployment.
- ML & Databases: built DNN pipelines for classifying card properties and tracking stock in SQLite.
- APIs: created RESTful APIs that allowed stock to be listed on e-commerce vendors seamlessly.
- Robotics: card-sorting arm prototype, attached a phone camera to an Arduino robot arm to allow automation of manually-intensive tasks of sorting new stock.
Education
Dissertation: "A Multi-Objective Evolutionary Search Strategy for Feature Selection in Machine Learning Models"
Developed a novel evolutionary algorithm approach for feature selection in ML customer churn classification for Vodafone Data Analytics. I self-sourced this dissertation in industry, it was not available as an option in the pre-prepared list of projects from the university. The work demonstrated significant improvements over traditional feature selection methods across multiple benchmark datasets, and is available as a pre-print.
🥈 Runner-up in the Operational Research Society's May Hicks Award
BSc Chemistry
Awards
I was awarded a runner-up award for my MSc Dissertation in Industry with Vodafone Data Analytics. In collaboration with stakeholders, I built an ML model to reduce customer churn. This was my first large ML project with real-world industry utility.
Awarded for services to the Department of Chemistry; I arranged surveys with students, enabling the staff committee get feedback from students and improve on NUS student satisfaction. I organised the focus groups and collated the results, which were enacted the following academic year.
Engagement
At the University of Trento in Italy, I presented research on behalf of Optrak that explores advanced search methodologies for efficiently tuning multi-phase parallel hyperheuristic optimisers with an extensive library of operators. Approaches included Bayesian Optimisation and a host of metaheuristics.
Guest lectures at several universities and colleges, promoting Operational Research as a discipline. This includes workshops arranged with the school, talks and online events. These are on my own initiative, and not affiliated with any of my employers. Schools visited: Richmond College, University of Edinburgh and Hertford College.
Article contributor to the Operational Research Society's semi-annual publication. As someone that works in industry, I think it's important to engage with OR practitioners and publicise the impact that the discipline can have.
Leading workshops for Ocado's analysts, including sessions on Agentic AI, practical ML and advanced analytics concepts. Currently, we are covering the Orange Book of ML by Carl McBride Ellis as our course content. The plan is to hold hackathons on key business problems, driving business value through upskilling of people in the department.
Timeline
New Role: Senior Data Scientist
Jul 2025Ocado Logistics
New Role: OR Software Developer
Sep 2023Optrak Vehicle Routing Software
Graduated: MSc with Distinction
Sep 2023University of Edinburgh
Started: MSc Operational Research
Sep 2022University of Edinburgh
New Role: Software Engineer
Oct 2020WeBuyAnyCard
Graduated: BSc Chemistry
Sep 2020University College London