Measurements in. Explainable models out.
_{Electrical & energy engineering student at EHTP · diagnostics for power grids and electrolyzers · ideas tested on the bench}

cascade-gnn: early warning for cascading failures in power grids, built on the PowerGraph (NeurIPS 2024) dataset.

• Problem: after an initial outage, will the grid fail to serve demand, or cascade into more branch trips?
• Approach: five graph neural network variants against XGBoost baselines; random, operating-condition and contingency-grouped splits with leakage checks; gradient × input and integrated-gradients edge attributions scored against the simulated cascade; a Streamlit demo.
• Result: v0.3 runs end to end with 14 passing tests in CI, Docker, and a claim ledger that separates what the code shows from what still needs the official PowerGraph runs. No benchmark score is claimed yet.
• Stack: Python · PyTorch · XGBoost · scikit-learn · Streamlit · Docker · GitHub Actions · LaTeX

A student competition project on fault diagnostics for PEM electrolyzers.

• Problem: spot faults in a PEM electrolyzer from its electrical current signals.
• Approach: MCSA-inspired current-signal features, an XGBoost classifier, and SHAP to show which features drive each prediction.
• Stack: Python · signal processing · XGBoost · SHAP

My TIPE project: a resonant wireless power transfer prototype, built and measured on the bench.

• Result: 23% efficiency at 34 kHz, with R² = 0.992 in a model comparison against the Yates reference.
• In progress: an OT / SCADA anomaly-detection dashboard.

_{↑ back to top}

• Hardware and code, together. My projects connect resonant power transfer, embedded acquisition, signal analysis, and Python-based modelling.
• Results I can defend. I use baselines, grouped splits, integrated gradients and SHAP to check what a model is really doing, and I write down which claims the evidence supports.
• A maths foundation and work on the bench. CPGE MP gave me the foundations. The WPT prototype and diagnostic projects give me a place to apply them, test things, and improve.

The internship I’m looking for has room for both a notebook and a workbench. I’d be glad to help with measurements, prototypes, data analysis, or diagnostic tools, and learn from the engineers around me.

_{↑ back to top}

Stay tuned: I’m cooking more, and this board fills up as the work ships.
_{Hit Follow on this profile to catch the next release.}

_{↑ back to top}

I was six when I opened my dad’s old PC. I wanted to know what was inside, and electronics became something I kept coming back to.

Now I study Génie Électrique et Énergétique at EHTP, after CPGE MP and the CNC. I’m interested in how measurements from real equipment can help us understand faults. That has led me to projects on power grids, electrolyzers, and wireless power transfer.

I want to help electrical systems waste less energy and catch problems earlier. That starts with understanding the hardware, getting useful measurements, and building models whose results people can make sense of. That's the kind of engineering I want to get good at.

_{↑ back to top}

_{↑ back to top}

_{↑ back to top}

I also enjoy competitive mathematics and metroidvania games. There’s usually another problem to get absorbed in. I draw around the things I work on too; the signals and scenes below are illustrations, and the project results are in the notes above.

_{↑ back to top}

I'm looking for a summer 2027 internship (1–2 months) in electrical and energy engineering, industrial AI, or OT cybersecurity. Morocco or abroad, on-site, hybrid, or remote. I work in Arabic, French and English. CV available on request.

_{Each button opens a pre-filled email in your mail app. Fill in the blanks and send.}

Akeanti | Electrical & energy engineering student, EHTP (GEE)
Background   CPGE MP · CNC
Looking for  Summer 2027 internship, 1–2 months
Fields       Electrical & energy engineering · industrial AI · OT security
Location     Morocco or abroad · on-site, hybrid or remote
Languages    Arabic (native) · French (fluent) · English (professional)
Highlights   Resonant WPT prototype: 23% efficiency at 34 kHz, R² = 0.992 vs Yates reference
             cascade-gnn: 5 GNN variants vs XGBoost on PowerGraph, 14 tests in CI, Docker
             PEM electrolyzer fault diagnostics: MCSA-inspired features, XGBoost, SHAP
Skills       Python, C++, Bash, PyTorch, PyTorch Geometric, XGBoost, SHAP, scikit-learn,
             ESP32, Arduino, ADS1115, SCADA, Nmap, Wireshark, Git, Linux, Docker, LaTeX
Keywords     power systems, grid resilience, fault diagnostics, hydrogen, PEM electrolyzer,
             wireless power transfer, signal processing, graph neural networks,
             explainable AI, industrial AI, OT security, SCADA, embedded systems
Contact      akeantie@gmail.com · github.com/akeanti · CV on request

akeantie@gmail.com  ·  cascade-gnn  ·  back to top ↑

_{Personal site, less formal: akeanti.xyz}