Technical Portfolio

Every project here is public, and every claim links to the code. Where a result is weak or a project is early-stage, it says so.

Fault Diagnosis in PMSM Motors — CNN on Wavelet Scalograms

#

github.com/mulhamfetna/pmsm-fault-diagnosis-cnn-scalogram · Python · MIT

Detecting inter-turn stator faults in permanent-magnet synchronous motors by converting raw sensor signals into continuous-wavelet scalograms and classifying them with a CNN.

Trained on the real KAIST motor dataset (DOI 10.17632/rgn5brrgrn.5) — not synthetic data.

Results, reported honestly:

The vibration result is excellent. The current-channel result is not, and the repository says so. The dataset contains only four healthy recordings, so the perfect vibration score may not generalize — that caveat is in the README, not buried in a footnote.

38 unit tests. This is the project I would want to be judged on: real data, a strong result, a weak result, and no hiding of either.

Merged Open-Source Contributions — OpenCV & OpenDR

#

Three pull requests accepted into major open-source projects. Two landed in OpenCV core, one of the most widely used computer-vision libraries in the world.

Both OpenCV fixes concern non-ASCII path and text handling — the class of bug that quietly breaks the library for everyone outside the English-speaking world, and that is easy to miss if you never work in Arabic.

Contributing to a library used by millions means most submissions are rejected. I have opened 30+ pull requests across OpenCV, opencv-python, Ultralytics, and OpenDR; these three were accepted.

WiFi as a Radar Replacement for Automotive SLAM

#

github.com/mulhamfetna/wifi-radar-slam · Python · AGPL-3.0 Archived on Zenodo — DOI 10.5281/zenodo.21247288

Can ambient WiFi substitute for radar in automotive localization and mapping? A simulation-first feasibility study using Sionna ray tracing, testing whether the channel-state information available from commodity WiFi carries enough structure for SLAM.

Status: active research, feasibility stage. The literature review and simulation infrastructure are complete; there are no results yet. I publish it as it develops rather than after the fact.

Trading Strategy Finder

#

github.com/mulhamfetna/trading-strategy-finder · Python · MIT

A backtesting engine comparing scalping, day, and intraday strategies on NQ futures, with a machine-learning filter layer over signal generation.

Frozen at tag v1.0.0 with reproducible results: $633.65 net profit, 54.5% win rate, 2.62 profit factor.

Modest, believable numbers on a small account — which is the point. The engineering (versioning, tests, reproducibility) is the deliverable here, not a claim of a trading edge.

OpenCode Presentations

#

github.com/mulhamfetna/opencode-presentations-skill · JavaScript · MIT · published on npm

A Marp-based presentation generator with 20 design styles and live preview. Small, finished, and actually shipped — a working tool anyone can install and use today.

Geopolymer Design System

#

github.com/mulhamfetna/geopolymer-design-system · Jupyter · Apache-2.0 · archived on Zenodo with a DOI

A Gradio application pairing a materials database with a RandomForest predictor and a Nelder–Mead optimizer, to design geopolymer mixes for nuclear-waste adsorption.

Caveat stated up front: the model is trained on synthetic data, not laboratory measurements. It is a design-exploration tool, and explicitly not a validated substitute for experimental work.

Embedded Systems — Ala’a Screens

#

Professional work, 2022–2023. Embedded systems design for P10 DMD LED display applications, including a basketball stadium scoreboard clock and a school-bus information display. Firmware and hardware, shipped and deployed.

Code: github.com/mulhamfetna · Contact: [email protected]