Master 2 student in Communicating Electronic Systems Design (CSEC) at the University of Lorraine (Metz, France). Before that I did a first Master's in Embedded Systems Electronics at USTHB, Algiers.
I like working where hardware meets software: digital design on FPGA in VHDL, firmware on microcontrollers, and the PC tools that talk to the board. Lately I've been spending more time on AI: hardware for AI inference (my FP8 project), how to evaluate a model honestly when the data is hard (my fraud detection project), and how language models actually see text (my French tokenizer, down to a C encoder on a Raspberry Pi).
🎯 Looking for a 4 to 6-month end-of-studies internship starting early 2027, in FPGA / digital design, embedded systems or embedded AI, anywhere in France.
| Project | What it is | Stack |
|---|---|---|
| FP8 MAC unit | Multiply-accumulate operator in FP8 (E4M3) for AI inference, synthesized on Cyclone IV (≈ 6 % of the FPGA) and verified with testbenches | VHDL · Quartus · Questa · GHDL |
| French BPE tokenizer | Byte-level BPE written from scratch and trained on French Wikipedia, with a pre-tokenization regex designed for French. The same text costs 27 % more tokens with GPT-4's tokenizer than with mine (11 % with GPT-4o's), and my French regex turned out to save only 2.45 %. A C99 encoder gives exactly the Python ids on every test document and runs at 14 MB/s on a Raspberry Pi 5; CI runs it under ASan and UBSan | Python · C99 · regex · tiktoken · SentencePiece · pytest · GitHub Actions |
| Fraud detection under an alert budget | Credit card fraud models (0.17 % fraud) evaluated the way a fraud team would use them: temporal split, alert budget, simulated cost in euros and bootstrap intervals. XGBoost catches 80 % of the frauds at a 0.1 % false positive rate; SMOTE brought no measurable gain, and a random split would have overstated PR-AUC by 0.07. 70 tests run in CI | Python · scikit-learn · XGBoost · imbalanced-learn · pytest · GitHub Actions |
| Qt6 UART API (PC ↔ FPGA) | C++ class that handles the serial link between a PC and a Basys3 board, tested on the real board, with QTest tests running in CI on virtual serial ports | C++17 · Qt 6 · QSerialPort · CMake · GitHub Actions |
| Smart House solar supply | Solar power chain for an Arduino house model (my part of a licence group project): PV sizing redone in Python, tested in CI and compared with multimeter measurements on the prototype | Arduino · TP4056 · XL6009 · Python · pytest · GitHub Actions |
| Face recognition attendance on Raspberry Pi | Time clock that recognises people with a webcam (Haar cascade + KNN) and logs arrivals in a daily CSV, with a Flask page to enrol users. I fixed the bugs from my report version and added 16 pytest tests running in CI | Python · OpenCV · scikit-learn · Flask · Raspberry Pi · pytest · GitHub Actions |
| Connected weather station (Home Assistant) | Web dashboard I wrote for a 10-student IoT lab: it reads the ESP32 sensors through the Home Assistant REST API and compares them with Open-Meteo. I also helped integrate the sensors. 13 tests on the unchanged code run in CI and caught one real bug | JavaScript · HTML/CSS · REST API · Home Assistant · ESP32 · node:test · GitHub Actions |
| Connected health monitor (ESP32 + Raspberry Pi 5) | My first Master's thesis: ECG, SpO2, temperature and air quality read by an ESP32 and sent over Wi-Fi to a Flask/SQLite server on a Pi 5, with a web dashboard and a Flutter app. Going back through the code, I found and fixed a bug that truncated the ECG to whole volts. Server tests, the firmware build and Flutter tests run in CI | ESP32 · Arduino C++ · Python · Flask · SQLite · Flutter · pytest · GitHub Actions |
| FPGA accelerator for molecular dynamics (in progress) | Hardware/software co-design: Python host + VHDL pipelines for force computation | VHDL · Python |
| Low-power LoRaWAN metering node | ESP32 firmware for voltage/current measurement, deep-sleep state machine, LoRa-E5 module (internship, CREDEG – Sonelgaz) | C/C++ · ESP32 · LoRaWAN · MQTT |
The projects without a link aren't published yet.
- AI / computer vision / NLP: OpenCV (Haar cascade face detection), scikit-learn (KNN, logistic regression), XGBoost, imbalanced-learn, model evaluation on imbalanced data (PR-AUC, bootstrap intervals, threshold choice), BPE tokenization (tiktoken, SentencePiece), neural network fundamentals, reduced-precision arithmetic (FP8) for inference, TensorFlow / PyTorch
- Digital design / FPGA: VHDL, Verilog, RTL design, testbenches, Intel Quartus Prime, Questa, Cyclone IV, Artix-7 (Basys 3)
- Embedded systems: C, C++, STM32, ESP32, Arduino, Raspberry Pi, FreeRTOS, UART, SPI, I2C, CAN, LoRaWAN
- Software & tools: Python, pytest, Qt 6, CMake, Git, GitHub Actions, Linux, MATLAB/Simulink, Flask, MQTT
- Web / IoT / mobile: JavaScript, HTML/CSS, Chart.js, REST APIs, Home Assistant, Flutter (Android)
English · French · Spanish · Arabic · Kabyle