Huddersfield, UK. Currently on railway hazard detection
Research Associate in computer vision at the University of Huddersfield, after ten years building and leading software teams.
I build detection models for railway and industrial inspection, then measure what happens to them outside the benchmark: another camera, worse weather, motion blur, defects that hardly appear in the data.
Citations, h-index and i10-index from the Google Scholar profile, as of 20 August 2026. The headline total is Scholar’s own de-duplicated figure, which is why it does not equal the sum of the per-paper counts below.
01 · About
I spent a decade as an engineer and engineering lead across Malaysia, Bangladesh and the UK, the last stretch as a principal engineer and acting CTO. Then I took an MSc in Artificial Intelligence at the University of Huddersfield and finished it with Distinction. I’ve been a Research Associate there since 2024.
My work sits between those two lives: YOLO-family detectors, CNN and transformer hybrids, applied to railway components, vehicles, construction sites and retinal images. The part I enjoy most is the unglamorous one: checking whether a result survives a change of camera, season or operating point.
02 · Research
Twelve papers, 2017 to 2026, across arXiv, an IEEE conference and peer-reviewed journals. h-index 11, i10-index 11.
310 cit.
240 cit.
121 cit.
107 cit.
29 cit.
24 cit.
12 cit.
3 cit.
11 cit.
5 cit.
109 cit.
17 cit.
03 · Applied
Projects where the constraint mattered as much as the model.
UKRI-funded · Jan–Mar 2025
+18%
detection coverage of under-represented hazard classes
Led an end-to-end research project generating synthetic railway hazard scenarios with Stable Diffusion, so detectors could learn classes that barely appear in real track footage.
Industrial inspection · 2023–24
97%
image-task efficiency, up from 92%
Object detection for pallet racking damage in a working warehouse setting, built to sit inside an existing compliance inspection workflow rather than replace it.
04 · Experience
Research Associate, Computer Vision
University of Huddersfield, UK
Synthetic generation of railway hazard scenarios; evaluation and fine-tuning pipelines for vision-language models; YOLOv5–YOLOv10 variants in real-time transport systems.
May 2024 – present
AI Engineer, LLM Systems
University of Huddersfield, UK
Designed a multi-agent chatbot system on LangChain and LangGraph, with contextual prompt modules and evaluation workflows for incremental fine-tuning.
May 2024 – Aug 2024
AI Engineer, Computer Vision
University of Huddersfield, UK
Detection software for pallet racking inspection; image-task efficiency raised from 92% to 97%.
Oct 2023 – Apr 2024
Principal Software Engineer
Welltravel, UK (remote)
Refactored the codebase for a 40% performance gain; led a cross-functional team of engineers and designers.
Sep 2022 – Sep 2023
Technical Lead / Acting CTO
Welltravel, Bangladesh
Owned the technology roadmap across three teams of about 20 people each, plus company-level architecture and data configuration.
Jul 2020 – Sep 2022
Senior Software Engineer
Welltravel, Bangladesh
Mentored and onboarded 15 engineers in Ruby on Rails; team delivery improved by 25%.
Nov 2017 – Jul 2020
Software Engineer
MyCash Online, Malaysia
Designed solutions for functionality, scalability and performance, and coordinated large-scale deployments.
Aug 2015 – Dec 2016
Junior Software Engineer
PixeLizard, Malaysia
Backend systems and APIs for client projects; the foundational years.
Jun 2014 – Jul 2015
05 · Interests
Three lines of work. The tools are the ones I use week to week, not a list of everything I have touched.
Object detection where the deployment distribution is not the training one: another camera, another season, motion blur, small and rare defects. Mostly railway components, vehicles and construction sites.
Generating the scenarios a dataset lacks, then testing whether synthetic-to-real transfer actually holds instead of assuming it.
Multi-agent tooling for research work: retrieval, contextual prompting, and evaluation loops with a human in the middle.
06 · Writing
07 · Contact
Email is fastest. Research collaboration, applied computer vision work, PhD supervision enquiries: all read.