Final-year electrical engineer at NED University — reproducing state-of-the-art deep-learning prognostics on real rotating machinery, with four merged PRs to ROOT and ROS 2 and a fully-funded research internship at Saint Mary's University, Halifax ahead.
My field has always been physical: motors, switchgear, microcontrollers. But the questions I keep coming back to aren't really electrical — they're statistical. How long before this bearing fails? Is this signal a fault, or just noise? What does this engine's healthy behaviour actually look like, and when does it start to drift?
So I've spent the last two years moving deeper into machine learning for physical systems — Bayesian uncertainty on dielectron data, sensor-fusion SLAM on mobile robots, and now an industrial-grade digital twin for my final-year project, modelled on the most recent prognostics literature.
"I'm not interested in ML as an abstract object. I want it to know my machines as well as my machines know themselves."
When I'm not building, I'm contributing to the open-source scientific tools I rely on — three merged PRs to ROOT and one to the ROS 2 controllers stack. In April 2026 I head to Saint Mary's University, Halifax on the MITACS Globalink Research Internship for multi-sensor SLAM on mobile robots — a competitive program funded by the Government of Canada.
Three classes of rotating machinery — gas turbines, steam turbines, and centrifugal pumps — each modelled with a physics-informed digital twin for real-time fault detection and degradation tracking. An MQTT–FastAPI sensor pipeline ingests live parameters; a Raspberry Pi dashboard surfaces equipment health and supports off-site experimentation via Tailscale. 50K+ labelled synthetic fault events drive the predictive-maintenance and remaining-useful-life models.
Faithfully reimplementing the best purely data-driven RUL predictor on N-CMAPSS, then extending it for the FYP's pumps and steam turbines.
Sun et al. report RMSE = 2.86 on DS02 after the smoothing step — the best purely data-driven result published, ahead of Deep Gaussian Processes (7.31), Transformer (6.29), and CNN-LSTM (6.66).
My reproduction matches the paper's PDT reconstruction error within tolerance; CNN tuning underway.
PyTorch · Keras · NumPy · pandas · scikit-learn · Matplotlib. Trained locally on a single GPU; pipeline scripted for reproducibility on the FYP Raspberry Pi for inference.
Two of the most important pieces of open-source infrastructure in my fields — multi-million-line codebases, multi-reviewer processes, tests that ship to thousands of downstream users.
Identified and fixed an incorrect residual calculation in the Chi2Test routine that affected goodness-of-fit comparisons on N-dimensional histograms — used across HEP analyses for hypothesis testing.
Refactored file-permission handling so ROOT's I/O subsystems correctly respect the OS-level umask. Improves multi-user reproducibility on shared compute clusters typical of physics labs.
View on GitHub ↗Diagnosed a floating-point rounding issue in GDML geometry serialization that subtly distorted exported detector geometries. Patched the export path to preserve double precision throughout.
View on GitHub ↗Fixed invalid YAML indentation in the joint_trajectory_controller docs example config and removed a stray trailing quote in a state-interface restriction bullet. Backported across three active ROS 2 distributions: humble jazzy kilted
From power-system protection in ETAP to ensemble ML on Pakistani agricultural data — work spanning the full electrical-engineering stack. Click any card to focus.
Not a buzzword cloud — these are the tools I've shipped real work with.
Competitively selected for Canada's flagship international research program — fully funded by the Government of Canada. Hosted at Saint Mary's University, Halifax for 12 weeks of multi-sensor fusion research applied to mobile-robot navigation and mapping.
Merit-based scholarship awarded for sustained academic performance at NED University of Engineering & Technology.
Presented paper: "Innovative Integration of Solar PV and Battery Storage Systems for Energy Efficiency."
Participant in advanced workshop on particle-physics detector technologies — silicon tracking, alignment, and reconstruction.
Open to graduate research positions, applied-ML collaborations, and industrial automation problems where physics, statistics, and embedded systems intersect. I read every email.