AI, Computer Vision & Scientific Computing
Researcher — applied AI, computer vision & scientific software
Where software meets science. My research turns hard problems into tools that can be tested, not just argued — across three strands: AI & computer vision, scientific software for astronomy, and applied machine learning.
Software meets science
I work where rigorous engineering meets open scientific questions. The method is constant: build a computational tool that lets a question be tested — a deep-learning model, a fast simulator, a reproducible pipeline — and ship it so others can use it. Three strands carry that method.
1 · AI & computer vision
A named co-inventor on the international patent WO 2021/198731 — an AI- and computer-vision-based method that uses UAV/drone imagery and deep learning to diagnose plant disease, nutrient deficiency and development in agricultural and horticultural crops.
2 · Scientific software & astronomy
The BSN Application — scientific software for photometric light-curve analysis of contact binary star systems, with an integrated MCMC panel that generates synthetic light curves 40× faster than PHOEBE, contributed to the international Binary Systems of South and North (BSN) project.
3 · Applied machine learning
Peer-reviewed machine learning for real engineering problems — air-cooled heat-sink optimization and thermal-management ANN models — and the DataStudio tooling for annotating and augmenting image and video datasets for computer-vision and NLP models.
AI & computer vision
A named co-inventor on an international AI / computer-vision patent (WO 2021/198731): UAV imagery and deep learning for plant-health diagnosis.
The question
Can a camera on a drone, plus deep learning, do what an agronomist does on foot — spot disease and nutrient deficiency early, across a whole field, at scale?
The patent
WO 2021/198731 A1 (PCT/IB2020/053083) — An AI-based method of agricultural and horticultural plants' physical characteristics and health diagnosing and development assessment. It uses UAVs/drones, cameras and deep-learning computer vision to diagnose plant disease and nutrient deficiency and to assess development. Filed 2020, published 2021; I am a named co-inventor. Read it on Google Patents.
Why it matters
Image-based diagnosis turns expensive, slow, expert-bound inspection into something a grower can run repeatedly and cheaply — the same computer-vision techniques transfer to medical imaging, industrial inspection and any domain where an expert eye is the bottleneck.
Scientific software & astronomy
The BSN Application — photometric light-curve analysis of contact binary stars with an MCMC panel, generating synthetic light curves 40× faster than PHOEBE.
The question
Contact binary stars share an envelope; their brightness rises and falls as they orbit. Fitting that light curve to recover the physical parameters is slow. Can it be made fast enough to explore?
The BSN Application
Scientific software for photometric light-curve analysis of contact binary systems, with an integrated MCMC panel for parameter estimation. It generates synthetic light curves 40× faster than PHOEBE, with an intuitive interface and full scientific-standard compliance — built within the international Binary Systems of South and North (BSN) project / Raderon's astronomy work.
Why it matters
A 40× speed-up changes what a researcher can do — from fitting one system at a time to sweeping parameter space and exploring populations. The same MCMC + fast-forward-model pattern recurs across scientific computing wherever an expensive simulator sits inside an inference loop.
Applied machine learning
Peer-reviewed ML for engineering — air-cooled heat-sink optimization and thermal ANN models — plus data-annotation and augmentation pipelines for vision and NLP.
The question
Engineering is full of expensive simulations and slow experiments. Where can a trained model replace the loop and still be trusted?
Peer-reviewed work
- Machine learning-based optimization of air-cooled heat sinks — Thermal Science and Engineering Progress 34, 101398 (2022); my most-cited paper.
- Prediction Accuracy of ANNs in Thermal Management Subject to Network Architectures — a study of how network architecture governs ANN accuracy in thermal-management prediction.
DataStudio
A tool to gather, annotate and augment image and video datasets for computer-vision and NLP models — sentiment analysis, entity linking and parsing pipelines — because models are only as good as the data that trains them.