Who am I

I’m a first-year PhD researcher at the Data Lab for Social Good in Cardiff Business School. My background is in civil and construction engineering, but most of my research has been about teaching computers to see and read: computer vision for safety on building sites, and more recently language models for risk assessment. Now I’m pointing those tools at a different kind of problem, which is how a changing climate is going to affect the way health and care services get delivered.

What my PhD is about

Wales floods. Roughly one in eight properties here is at some risk of it, and the climate signal points to wetter winters and higher seas. Hospitals and clinics can’t simply shut their doors when the water rises, which makes the failure modes the interesting part: how services break, who loses access, and for how long.

That is harder to pin down than it sounds. Flood maps tell you which buildings get wet. They don’t tell you that a flooded substation three streets away has knocked out a dialysis unit, or that staff couldn’t reach a ward because the roads were under water. A hospital can stay completely dry and still stop delivering care.

My research builds the layer that is missing. The first part maps the mechanisms through which floods disrupt healthcare, turning scattered evidence from reports, academic papers, news, and the Senedd record into a structured picture of what actually fails. From there the project moves toward scoring which Welsh facilities are most exposed, projecting how that exposure shifts under different climate futures, and helping planners decide where to put resources before the next flood rather than during it. A good chunk of this runs on a text-mining pipeline I’m building, which uses large language models to pull coded events out of messy documents at a scale no single reader could keep up with.

The work is grounded in Wales, but the methods are meant to travel.

Supervision team

Funding and partners

The PhD is funded by an ESRC studentship through the Wales Graduate School for the Social Sciences (WGSSS). It is shaped by working with people who handle this problem directly: Public Health Wales, the Future Generations Commissioner for Wales, Natural Resources Wales, and the Office for National Statistics. Their experience has already changed how I framed the project, including a shift away from real-time response toward longer-term preparedness planning, because the data you would need during a live incident mostly isn’t there.

Before Cardiff

I did both my BSc in Civil Engineering and my MSc in Construction Engineering and Management at Amirkabir University of Technology in Tehran, finishing second in my master’s cohort. My thesis built a computer vision system to check whether construction workers were wearing their protective equipment, combining object detection, image classification, and fuzzy logic. After that I spent two years as a visiting researcher at Purdue University, working with Dr Sogand Hasanzadeh on machine learning and language models for assessing the risk of vehicles intruding into roadwork zones.

What I work with

Most of my day to day is in Python: PyTorch and TensorFlow for deep learning, the usual data and ML stack, and increasingly the tooling that sits around large language models. I also work in R, mainly for forecasting and statistics. My interests sit where machine learning, computer vision, NLP, forecasting, operations research, and optimisation overlap, which is a roundabout way of saying I like problems that need both a model and a reason to trust what it tells you.

A quick rundown:

  • Languages and tools: Python, R, PyTorch, TensorFlow, Git
  • Methods: machine learning, deep learning, computer vision, NLP and large language models, forecasting, statistics, operations research, optimisation, simulation
  • Spoken languages: Persian (native), English (TOEFL 110)

Where I’m headed

After the PhD I want to work as an applied scientist, machine learning engineer, or data scientist, ideally in London. So alongside the research I’m deliberately building the engineering side of the job: putting models into pipelines other people can actually run, instead of notebooks that only work once on my own laptop.